r16945: Sync trunk -> 3.0 for 3.0.24 code. Still need

to do the upper layer directories but this is what
everyone is waiting for....

Jeremy.
(This used to be commit 9dafb7f48ca3e7af956b0a7d1720c2546fc4cfb8)

16 files changed, 3553 insertions, 2578 deletions

diff --git a/source3/tdb/dump.c b/source3/tdb/dump.c
new file mode 100644
index 0000000000..577f23aac6
--- /dev/null
+++ b/source3/tdb/dump.c
@@ -0,0 +1,135 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+static tdb_off_t tdb_dump_record(struct tdb_context *tdb, tdb_off_t offset)
+{
+	struct list_struct rec;
+	tdb_off_t tailer_ofs, tailer;
+
+	if (tdb->methods->tdb_read(tdb, offset, (char *)&rec, 
+				   sizeof(rec), DOCONV()) == -1) {
+		printf("ERROR: failed to read record at %u\n", offset);
+		return 0;
+	}
+
+	printf(" rec: offset=0x%08x next=0x%08x rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
+	       offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
+
+	tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off_t);
+
+	if (tdb_ofs_read(tdb, tailer_ofs, &tailer) == -1) {
+		printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
+		return rec.next;
+	}
+
+	if (tailer != rec.rec_len + sizeof(rec)) {
+		printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
+				(unsigned int)tailer, (unsigned int)(rec.rec_len + sizeof(rec)));
+	}
+	return rec.next;
+}
+
+static int tdb_dump_chain(struct tdb_context *tdb, int i)
+{
+	tdb_off_t rec_ptr, top;
+
+	top = TDB_HASH_TOP(i);
+
+	if (tdb_lock(tdb, i, F_WRLCK) != 0)
+		return -1;
+
+	if (tdb_ofs_read(tdb, top, &rec_ptr) == -1)
+		return tdb_unlock(tdb, i, F_WRLCK);
+
+	if (rec_ptr)
+		printf("hash=%d\n", i);
+
+	while (rec_ptr) {
+		rec_ptr = tdb_dump_record(tdb, rec_ptr);
+	}
+
+	return tdb_unlock(tdb, i, F_WRLCK);
+}
+
+void tdb_dump_all(struct tdb_context *tdb)
+{
+	int i;
+	for (i=0;i<tdb->header.hash_size;i++) {
+		tdb_dump_chain(tdb, i);
+	}
+	printf("freelist:\n");
+	tdb_dump_chain(tdb, -1);
+}
+
+int tdb_printfreelist(struct tdb_context *tdb)
+{
+	int ret;
+	long total_free = 0;
+	tdb_off_t offset, rec_ptr;
+	struct list_struct rec;
+
+	if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
+		return ret;
+
+	offset = FREELIST_TOP;
+
+	/* read in the freelist top */
+	if (tdb_ofs_read(tdb, offset, &rec_ptr) == -1) {
+		tdb_unlock(tdb, -1, F_WRLCK);
+		return 0;
+	}
+
+	printf("freelist top=[0x%08x]\n", rec_ptr );
+	while (rec_ptr) {
+		if (tdb->methods->tdb_read(tdb, rec_ptr, (char *)&rec, 
+					   sizeof(rec), DOCONV()) == -1) {
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return -1;
+		}
+
+		if (rec.magic != TDB_FREE_MAGIC) {
+			printf("bad magic 0x%08x in free list\n", rec.magic);
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return -1;
+		}
+
+		printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)] (end = 0x%08x)\n", 
+		       rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + rec.rec_len);
+		total_free += rec.rec_len;
+
+		/* move to the next record */
+		rec_ptr = rec.next;
+	}
+	printf("total rec_len = [0x%08x (%d)]\n", (int)total_free, 
+               (int)total_free);
+
+	return tdb_unlock(tdb, -1, F_WRLCK);
+}
+
diff --git a/source3/tdb/error.c b/source3/tdb/error.c
new file mode 100644
index 0000000000..4cf33a29ab
--- /dev/null
+++ b/source3/tdb/error.c
@@ -0,0 +1,58 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+{
+	return tdb->ecode;
+}
+
+static struct tdb_errname {
+	enum TDB_ERROR ecode; const char *estring;
+} emap[] = { {TDB_SUCCESS, "Success"},
+	     {TDB_ERR_CORRUPT, "Corrupt database"},
+	     {TDB_ERR_IO, "IO Error"},
+	     {TDB_ERR_LOCK, "Locking error"},
+	     {TDB_ERR_OOM, "Out of memory"},
+	     {TDB_ERR_EXISTS, "Record exists"},
+	     {TDB_ERR_NOLOCK, "Lock exists on other keys"},
+	     {TDB_ERR_EINVAL, "Invalid parameter"},
+	     {TDB_ERR_NOEXIST, "Record does not exist"},
+	     {TDB_ERR_RDONLY, "write not permitted"} };
+
+/* Error string for the last tdb error */
+const char *tdb_errorstr(struct tdb_context *tdb)
+{
+	u32 i;
+	for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
+		if (tdb->ecode == emap[i].ecode)
+			return emap[i].estring;
+	return "Invalid error code";
+}
+
diff --git a/source3/tdb/freelist.c b/source3/tdb/freelist.c
new file mode 100644
index 0000000000..3483751164
--- /dev/null
+++ b/source3/tdb/freelist.c
@@ -0,0 +1,331 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+/* read a freelist record and check for simple errors */
+static int rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct list_struct *rec)
+{
+	if (tdb->methods->tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
+		return -1;
+
+	if (rec->magic == TDB_MAGIC) {
+		/* this happens when a app is showdown while deleting a record - we should
+		   not completely fail when this happens */
+		TDB_LOG((tdb, 0,"rec_free_read non-free magic 0x%x at offset=%d - fixing\n", 
+			 rec->magic, off));
+		rec->magic = TDB_FREE_MAGIC;
+		if (tdb->methods->tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
+			return -1;
+	}
+
+	if (rec->magic != TDB_FREE_MAGIC) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n", 
+			   rec->magic, off));
+		return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+	}
+	if (tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
+		return -1;
+	return 0;
+}
+
+
+
+/* Remove an element from the freelist.  Must have alloc lock. */
+static int remove_from_freelist(struct tdb_context *tdb, tdb_off_t off, tdb_off_t next)
+{
+	tdb_off_t last_ptr, i;
+
+	/* read in the freelist top */
+	last_ptr = FREELIST_TOP;
+	while (tdb_ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
+		if (i == off) {
+			/* We've found it! */
+			return tdb_ofs_write(tdb, last_ptr, &next);
+		}
+		/* Follow chain (next offset is at start of record) */
+		last_ptr = i;
+	}
+	TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
+	return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+}
+
+
+/* update a record tailer (must hold allocation lock) */
+static int update_tailer(struct tdb_context *tdb, tdb_off_t offset,
+			 const struct list_struct *rec)
+{
+	tdb_off_t totalsize;
+
+	/* Offset of tailer from record header */
+	totalsize = sizeof(*rec) + rec->rec_len;
+	return tdb_ofs_write(tdb, offset + totalsize - sizeof(tdb_off_t),
+			 &totalsize);
+}
+
+/* Add an element into the freelist. Merge adjacent records if
+   neccessary. */
+int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
+{
+	tdb_off_t right, left;
+
+	/* Allocation and tailer lock */
+	if (tdb_lock(tdb, -1, F_WRLCK) != 0)
+		return -1;
+
+	/* set an initial tailer, so if we fail we don't leave a bogus record */
+	if (update_tailer(tdb, offset, rec) != 0) {
+		TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
+		goto fail;
+	}
+
+	/* Look right first (I'm an Australian, dammit) */
+	right = offset + sizeof(*rec) + rec->rec_len;
+	if (right + sizeof(*rec) <= tdb->map_size) {
+		struct list_struct r;
+
+		if (tdb->methods->tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
+			TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
+			goto left;
+		}
+
+		/* If it's free, expand to include it. */
+		if (r.magic == TDB_FREE_MAGIC) {
+			if (remove_from_freelist(tdb, right, r.next) == -1) {
+				TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
+				goto left;
+			}
+			rec->rec_len += sizeof(r) + r.rec_len;
+		}
+	}
+
+left:
+	/* Look left */
+	left = offset - sizeof(tdb_off_t);
+	if (left > TDB_DATA_START(tdb->header.hash_size)) {
+		struct list_struct l;
+		tdb_off_t leftsize;
+		
+		/* Read in tailer and jump back to header */
+		if (tdb_ofs_read(tdb, left, &leftsize) == -1) {
+			TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
+			goto update;
+		}
+
+		/* it could be uninitialised data */
+		if (leftsize == 0 || leftsize == TDB_PAD_U32) {
+			goto update;
+		}
+
+		left = offset - leftsize;
+
+		/* Now read in record */
+		if (tdb->methods->tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
+			TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
+			goto update;
+		}
+
+		/* If it's free, expand to include it. */
+		if (l.magic == TDB_FREE_MAGIC) {
+			if (remove_from_freelist(tdb, left, l.next) == -1) {
+				TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
+				goto update;
+			} else {
+				offset = left;
+				rec->rec_len += leftsize;
+			}
+		}
+	}
+
+update:
+	if (update_tailer(tdb, offset, rec) == -1) {
+		TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
+		goto fail;
+	}
+
+	/* Now, prepend to free list */
+	rec->magic = TDB_FREE_MAGIC;
+
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
+	    tdb_rec_write(tdb, offset, rec) == -1 ||
+	    tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+		TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
+		goto fail;
+	}
+
+	/* And we're done. */
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return 0;
+
+ fail:
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return -1;
+}
+
+
+/* 
+   the core of tdb_allocate - called when we have decided which
+   free list entry to use
+ */
+static tdb_off_t tdb_allocate_ofs(struct tdb_context *tdb, tdb_len_t length, tdb_off_t rec_ptr,
+				struct list_struct *rec, tdb_off_t last_ptr)
+{
+	struct list_struct newrec;
+	tdb_off_t newrec_ptr;
+
+	memset(&newrec, '\0', sizeof(newrec));
+
+	/* found it - now possibly split it up  */
+	if (rec->rec_len > length + MIN_REC_SIZE) {
+		/* Length of left piece */
+		length = TDB_ALIGN(length, TDB_ALIGNMENT);
+		
+		/* Right piece to go on free list */
+		newrec.rec_len = rec->rec_len - (sizeof(*rec) + length);
+		newrec_ptr = rec_ptr + sizeof(*rec) + length;
+		
+		/* And left record is shortened */
+		rec->rec_len = length;
+	} else {
+		newrec_ptr = 0;
+	}
+	
+	/* Remove allocated record from the free list */
+	if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1) {
+		return 0;
+	}
+	
+	/* Update header: do this before we drop alloc
+	   lock, otherwise tdb_free() might try to
+	   merge with us, thinking we're free.
+	   (Thanks Jeremy Allison). */
+	rec->magic = TDB_MAGIC;
+	if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+		return 0;
+	}
+	
+	/* Did we create new block? */
+	if (newrec_ptr) {
+		/* Update allocated record tailer (we
+		   shortened it). */
+		if (update_tailer(tdb, rec_ptr, rec) == -1) {
+			return 0;
+		}
+		
+		/* Free new record */
+		if (tdb_free(tdb, newrec_ptr, &newrec) == -1) {
+			return 0;
+		}
+	}
+	
+	/* all done - return the new record offset */
+	return rec_ptr;
+}
+
+/* allocate some space from the free list. The offset returned points
+   to a unconnected list_struct within the database with room for at
+   least length bytes of total data
+
+   0 is returned if the space could not be allocated
+ */
+tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct list_struct *rec)
+{
+	tdb_off_t rec_ptr, last_ptr, newrec_ptr;
+	struct {
+		tdb_off_t rec_ptr, last_ptr;
+		tdb_len_t rec_len;
+	} bestfit;
+
+	if (tdb_lock(tdb, -1, F_WRLCK) == -1)
+		return 0;
+
+	/* Extra bytes required for tailer */
+	length += sizeof(tdb_off_t);
+
+ again:
+	last_ptr = FREELIST_TOP;
+
+	/* read in the freelist top */
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
+		goto fail;
+
+	bestfit.rec_ptr = 0;
+	bestfit.last_ptr = 0;
+	bestfit.rec_len = 0;
+
+	/* 
+	   this is a best fit allocation strategy. Originally we used
+	   a first fit strategy, but it suffered from massive fragmentation
+	   issues when faced with a slowly increasing record size.
+	 */
+	while (rec_ptr) {
+		if (rec_free_read(tdb, rec_ptr, rec) == -1) {
+			goto fail;
+		}
+
+		if (rec->rec_len >= length) {
+			if (bestfit.rec_ptr == 0 ||
+			    rec->rec_len < bestfit.rec_len) {
+				bestfit.rec_len = rec->rec_len;
+				bestfit.rec_ptr = rec_ptr;
+				bestfit.last_ptr = last_ptr;
+				/* consider a fit to be good enough if
+				   we aren't wasting more than half
+				   the space */
+				if (bestfit.rec_len < 2*length) {
+					break;
+				}
+			}
+		}
+
+		/* move to the next record */
+		last_ptr = rec_ptr;
+		rec_ptr = rec->next;
+	}
+
+	if (bestfit.rec_ptr != 0) {
+		if (rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
+			goto fail;
+		}
+
+		newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr, rec, bestfit.last_ptr);
+		tdb_unlock(tdb, -1, F_WRLCK);
+		return newrec_ptr;
+	}
+
+	/* we didn't find enough space. See if we can expand the
+	   database and if we can then try again */
+	if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
+		goto again;
+ fail:
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return 0;
+}
+
diff --git a/source3/tdb/io.c b/source3/tdb/io.c
new file mode 100644
index 0000000000..ac6f4cdf57
--- /dev/null
+++ b/source3/tdb/io.c
@@ -0,0 +1,412 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+
+#include "tdb_private.h"
+
+#ifndef HAVE_PREAD
+ static ssize_t pread(int fd, void *buf, size_t count, off_t offset)
+{
+	if (lseek(fd, offset, SEEK_SET) != offset) {
+		errno = EIO;
+		return -1;
+	}
+	return read(fd, buf, count);
+}
+#endif
+
+#ifndef HAVE_PWRITE
+ static ssize_t pwrite(int fd, const void *buf, size_t count, off_t offset)
+{
+	if (lseek(fd, offset, SEEK_SET) != offset) {
+		errno = EIO;
+		return -1;
+	}
+	return write(fd, buf, count);
+}
+#endif
+
+/* check for an out of bounds access - if it is out of bounds then
+   see if the database has been expanded by someone else and expand
+   if necessary 
+   note that "len" is the minimum length needed for the db
+*/
+static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
+{
+	struct stat st;
+	if (len <= tdb->map_size)
+		return 0;
+	if (tdb->flags & TDB_INTERNAL) {
+		if (!probe) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n",
+				 (int)len, (int)tdb->map_size));
+		}
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	}
+
+	if (fstat(tdb->fd, &st) == -1) {
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	}
+
+	if (st.st_size < (size_t)len) {
+		if (!probe) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
+				 (int)len, (int)st.st_size));
+		}
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	}
+
+	/* Unmap, update size, remap */
+	if (tdb_munmap(tdb) == -1)
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	tdb->map_size = st.st_size;
+	tdb_mmap(tdb);
+	return 0;
+}
+
+/* write a lump of data at a specified offset */
+static int tdb_write(struct tdb_context *tdb, tdb_off_t off, 
+		     const void *buf, tdb_len_t len)
+{
+	if (len == 0) {
+		return 0;
+	}
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0)
+		return -1;
+
+	if (tdb->map_ptr) {
+		memcpy(off + (char *)tdb->map_ptr, buf, len);
+	} else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_IO;
+		TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
+			   off, len, strerror(errno)));
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	}
+	return 0;
+}
+
+/* Endian conversion: we only ever deal with 4 byte quantities */
+void *tdb_convert(void *buf, u32 size)
+{
+	u32 i, *p = buf;
+	for (i = 0; i < size / 4; i++)
+		p[i] = TDB_BYTEREV(p[i]);
+	return buf;
+}
+
+
+/* read a lump of data at a specified offset, maybe convert */
+static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf, 
+		    tdb_len_t len, int cv)
+{
+	if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0) {
+		return -1;
+	}
+
+	if (tdb->map_ptr) {
+		memcpy(buf, off + (char *)tdb->map_ptr, len);
+	} else {
+		ssize_t ret = pread(tdb->fd, buf, len, off);
+		if (ret != (ssize_t)len) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d ret=%d (%s) map_size=%d\n",
+				 off, len, (int)ret, strerror(errno), (int)tdb->map_size));
+			return TDB_ERRCODE(TDB_ERR_IO, -1);
+		}
+	}
+	if (cv) {
+		tdb_convert(buf, len);
+	}
+	return 0;
+}
+
+
+
+/*
+  do an unlocked scan of the hash table heads to find the next non-zero head. The value
+  will then be confirmed with the lock held
+*/		
+static void tdb_next_hash_chain(struct tdb_context *tdb, u32 *chain)
+{
+	u32 h = *chain;
+	if (tdb->map_ptr) {
+		for (;h < tdb->header.hash_size;h++) {
+			if (0 != *(u32 *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
+				break;
+			}
+		}
+	} else {
+		u32 off=0;
+		for (;h < tdb->header.hash_size;h++) {
+			if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
+				break;
+			}
+		}
+	}
+	(*chain) = h;
+}
+
+
+int tdb_munmap(struct tdb_context *tdb)
+{
+	if (tdb->flags & TDB_INTERNAL)
+		return 0;
+
+#ifdef HAVE_MMAP
+	if (tdb->map_ptr) {
+		int ret = munmap(tdb->map_ptr, tdb->map_size);
+		if (ret != 0)
+			return ret;
+	}
+#endif
+	tdb->map_ptr = NULL;
+	return 0;
+}
+
+void tdb_mmap(struct tdb_context *tdb)
+{
+	if (tdb->flags & TDB_INTERNAL)
+		return;
+
+#ifdef HAVE_MMAP
+	if (!(tdb->flags & TDB_NOMMAP)) {
+		tdb->map_ptr = mmap(NULL, tdb->map_size, 
+				    PROT_READ|(tdb->read_only? 0:PROT_WRITE), 
+				    MAP_SHARED|MAP_FILE, tdb->fd, 0);
+
+		/*
+		 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
+		 */
+
+		if (tdb->map_ptr == MAP_FAILED) {
+			tdb->map_ptr = NULL;
+			TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n", 
+				 tdb->map_size, strerror(errno)));
+		}
+	} else {
+		tdb->map_ptr = NULL;
+	}
+#else
+	tdb->map_ptr = NULL;
+#endif
+}
+
+/* expand a file.  we prefer to use ftruncate, as that is what posix
+  says to use for mmap expansion */
+static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
+{
+	char buf[1024];
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	if (ftruncate(tdb->fd, size+addition) == -1) {
+		char b = 0;
+		if (pwrite(tdb->fd,  &b, 1, (size+addition) - 1) != 1) {
+			TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n", 
+				 size+addition, strerror(errno)));
+			return -1;
+		}
+	}
+
+	/* now fill the file with something. This ensures that the
+	   file isn't sparse, which would be very bad if we ran out of
+	   disk. This must be done with write, not via mmap */
+	memset(buf, TDB_PAD_BYTE, sizeof(buf));
+	while (addition) {
+		int n = addition>sizeof(buf)?sizeof(buf):addition;
+		int ret = pwrite(tdb->fd, buf, n, size);
+		if (ret != n) {
+			TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n", 
+				   n, strerror(errno)));
+			return -1;
+		}
+		addition -= n;
+		size += n;
+	}
+	return 0;
+}
+
+
+/* expand the database at least size bytes by expanding the underlying
+   file and doing the mmap again if necessary */
+int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
+{
+	struct list_struct rec;
+	tdb_off_t offset;
+
+	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+		TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
+		return -1;
+	}
+
+	/* must know about any previous expansions by another process */
+	tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+
+	/* always make room for at least 10 more records, and round
+           the database up to a multiple of the page size */
+	size = TDB_ALIGN(tdb->map_size + size*10, tdb->page_size) - tdb->map_size;
+
+	if (!(tdb->flags & TDB_INTERNAL))
+		tdb_munmap(tdb);
+
+	/*
+	 * We must ensure the file is unmapped before doing this
+	 * to ensure consistency with systems like OpenBSD where
+	 * writes and mmaps are not consistent.
+	 */
+
+	/* expand the file itself */
+	if (!(tdb->flags & TDB_INTERNAL)) {
+		if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
+			goto fail;
+	}
+
+	tdb->map_size += size;
+
+	if (tdb->flags & TDB_INTERNAL) {
+		char *new_map_ptr = realloc(tdb->map_ptr, tdb->map_size);
+		if (!new_map_ptr) {
+			tdb->map_size -= size;
+			goto fail;
+		}
+		tdb->map_ptr = new_map_ptr;
+	} else {
+		/*
+		 * We must ensure the file is remapped before adding the space
+		 * to ensure consistency with systems like OpenBSD where
+		 * writes and mmaps are not consistent.
+		 */
+
+		/* We're ok if the mmap fails as we'll fallback to read/write */
+		tdb_mmap(tdb);
+	}
+
+	/* form a new freelist record */
+	memset(&rec,'\0',sizeof(rec));
+	rec.rec_len = size - sizeof(rec);
+
+	/* link it into the free list */
+	offset = tdb->map_size - size;
+	if (tdb_free(tdb, offset, &rec) == -1)
+		goto fail;
+
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return 0;
+ fail:
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return -1;
+}
+
+/* read/write a tdb_off_t */
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+	return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
+}
+
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+	tdb_off_t off = *d;
+	return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
+}
+
+
+/* read a lump of data, allocating the space for it */
+char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
+{
+	char *buf;
+
+	/* some systems don't like zero length malloc */
+	if (len == 0) {
+		len = 1;
+	}
+
+	if (!(buf = malloc(len))) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_OOM;
+		TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
+			   len, strerror(errno)));
+		return TDB_ERRCODE(TDB_ERR_OOM, buf);
+	}
+	if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
+		SAFE_FREE(buf);
+		return NULL;
+	}
+	return buf;
+}
+
+/* read/write a record */
+int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
+{
+	if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
+		return -1;
+	if (TDB_BAD_MAGIC(rec)) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, 0,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
+		return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+	}
+	return tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0);
+}
+
+int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
+{
+	struct list_struct r = *rec;
+	return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
+}
+
+static const struct tdb_methods io_methods = {
+	tdb_read,
+	tdb_write,
+	tdb_next_hash_chain,
+	tdb_oob,
+	tdb_expand_file,
+	tdb_brlock
+};
+
+/*
+  initialise the default methods table
+*/
+void tdb_io_init(struct tdb_context *tdb)
+{
+	tdb->methods = &io_methods;
+}
diff --git a/source3/tdb/lock.c b/source3/tdb/lock.c
new file mode 100644
index 0000000000..a96d77a229
--- /dev/null
+++ b/source3/tdb/lock.c
@@ -0,0 +1,276 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+/* a byte range locking function - return 0 on success
+   this functions locks/unlocks 1 byte at the specified offset.
+
+   On error, errno is also set so that errors are passed back properly
+   through tdb_open(). 
+
+   note that a len of zero means lock to end of file
+*/
+int tdb_brlock_len(struct tdb_context *tdb, tdb_off_t offset, 
+		   int rw_type, int lck_type, int probe, size_t len)
+{
+	struct flock fl;
+	int ret;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	fl.l_type = rw_type;
+	fl.l_whence = SEEK_SET;
+	fl.l_start = offset;
+	fl.l_len = len;
+	fl.l_pid = 0;
+
+	do {
+		ret = fcntl(tdb->fd,lck_type,&fl);
+	} while (ret == -1 && errno == EINTR);
+
+	if (ret == -1) {
+		/* Generic lock error. errno set by fcntl.
+		 * EAGAIN is an expected return from non-blocking
+		 * locks. */
+		if (!probe && lck_type != F_SETLK) {
+			/* Ensure error code is set for log fun to examine. */
+			tdb->ecode = TDB_ERR_LOCK;
+			TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n", 
+				 tdb->fd, offset, rw_type, lck_type, (int)len));
+		}
+		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+	}
+	return 0;
+}
+
+
+/*
+  upgrade a read lock to a write lock. This needs to be handled in a
+  special way as some OSes (such as solaris) have too conservative
+  deadlock detection and claim a deadlock when progress can be
+  made. For those OSes we may loop for a while.  
+*/
+int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
+{
+	int count = 1000;
+	while (count--) {
+		struct timeval tv;
+		if (tdb_brlock_len(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
+			return 0;
+		}
+		if (errno != EDEADLK) {
+			break;
+		}
+		/* sleep for as short a time as we can - more portable than usleep() */
+		tv.tv_sec = 0;
+		tv.tv_usec = 1;
+		select(0, NULL, NULL, NULL, &tv);
+	}
+	TDB_LOG((tdb, 5,"tdb_brlock_upgrade failed at offset %d\n", offset));
+	return -1;
+}
+
+
+/* a byte range locking function - return 0 on success
+   this functions locks/unlocks 1 byte at the specified offset.
+
+   On error, errno is also set so that errors are passed back properly
+   through tdb_open(). */
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, 
+	       int rw_type, int lck_type, int probe)
+{
+	return tdb_brlock_len(tdb, offset, rw_type, lck_type, probe, 1);
+}
+
+/* lock a list in the database. list -1 is the alloc list */
+int tdb_lock(struct tdb_context *tdb, int list, int ltype)
+{
+	if (list < -1 || list >= (int)tdb->header.hash_size) {
+		TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n", 
+			   list, ltype));
+		return -1;
+	}
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
+	/* Since fcntl locks don't nest, we do a lock for the first one,
+	   and simply bump the count for future ones */
+	if (tdb->locked[list+1].count == 0) {
+		if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
+			TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n", 
+				 list, ltype, strerror(errno)));
+			return -1;
+		}
+		tdb->locked[list+1].ltype = ltype;
+		tdb->num_locks++;
+	}
+	tdb->locked[list+1].count++;
+	return 0;
+}
+
+/* unlock the database: returns void because it's too late for errors. */
+	/* changed to return int it may be interesting to know there
+	   has been an error  --simo */
+int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
+{
+	int ret = -1;
+
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
+	/* Sanity checks */
+	if (list < -1 || list >= (int)tdb->header.hash_size) {
+		TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
+		return ret;
+	}
+
+	if (tdb->locked[list+1].count==0) {
+		TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
+		return ret;
+	}
+
+	if (tdb->locked[list+1].count == 1) {
+		/* Down to last nested lock: unlock underneath */
+		ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
+		tdb->num_locks--;
+	} else {
+		ret = 0;
+	}
+	tdb->locked[list+1].count--;
+
+	if (ret)
+		TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n")); 
+	return ret;
+}
+
+
+
+/* lock/unlock entire database */
+int tdb_lockall(struct tdb_context *tdb)
+{
+	u32 i;
+
+	/* There are no locks on read-only dbs */
+	if (tdb->read_only || tdb->traverse_read)
+		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+	for (i = 0; i < tdb->header.hash_size; i++) 
+		if (tdb_lock(tdb, i, F_WRLCK))
+			break;
+
+	/* If error, release locks we have... */
+	if (i < tdb->header.hash_size) {
+		u32 j;
+
+		for ( j = 0; j < i; j++)
+			tdb_unlock(tdb, j, F_WRLCK);
+		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
+	}
+
+	return 0;
+}
+void tdb_unlockall(struct tdb_context *tdb)
+{
+	u32 i;
+	for (i=0; i < tdb->header.hash_size; i++)
+		tdb_unlock(tdb, i, F_WRLCK);
+}
+
+/* lock/unlock one hash chain. This is meant to be used to reduce
+   contention - it cannot guarantee how many records will be locked */
+int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
+{
+	return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
+{
+	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+	return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+
+
+/* record lock stops delete underneath */
+int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
+}
+
+/*
+  Write locks override our own fcntl readlocks, so check it here.
+  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+  an error to fail to get the lock here.
+*/
+int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	struct tdb_traverse_lock *i;
+	for (i = &tdb->travlocks; i; i = i->next)
+		if (i->off == off)
+			return -1;
+	return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
+}
+
+/*
+  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+  an error to fail to get the lock here.
+*/
+int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
+}
+
+/* fcntl locks don't stack: avoid unlocking someone else's */
+int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	struct tdb_traverse_lock *i;
+	u32 count = 0;
+
+	if (off == 0)
+		return 0;
+	for (i = &tdb->travlocks; i; i = i->next)
+		if (i->off == off)
+			count++;
+	return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
+}
diff --git a/source3/tdb/open.c b/source3/tdb/open.c
new file mode 100644
index 0000000000..cb4a9cd994
--- /dev/null
+++ b/source3/tdb/open.c
@@ -0,0 +1,453 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
+static struct tdb_context *tdbs = NULL;
+
+
+/* This is based on the hash algorithm from gdbm */
+static unsigned int default_tdb_hash(TDB_DATA *key)
+{
+	u32 value;	/* Used to compute the hash value.  */
+	u32   i;	/* Used to cycle through random values. */
+
+	/* Set the initial value from the key size. */
+	for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
+		value = (value + (key->dptr[i] << (i*5 % 24)));
+
+	return (1103515243 * value + 12345);  
+}
+
+
+/* initialise a new database with a specified hash size */
+static int tdb_new_database(struct tdb_context *tdb, int hash_size)
+{
+	struct tdb_header *newdb;
+	int size, ret = -1;
+
+	/* We make it up in memory, then write it out if not internal */
+	size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
+	if (!(newdb = calloc(size, 1)))
+		return TDB_ERRCODE(TDB_ERR_OOM, -1);
+
+	/* Fill in the header */
+	newdb->version = TDB_VERSION;
+	newdb->hash_size = hash_size;
+	if (tdb->flags & TDB_INTERNAL) {
+		tdb->map_size = size;
+		tdb->map_ptr = (char *)newdb;
+		memcpy(&tdb->header, newdb, sizeof(tdb->header));
+		/* Convert the `ondisk' version if asked. */
+		CONVERT(*newdb);
+		return 0;
+	}
+	if (lseek(tdb->fd, 0, SEEK_SET) == -1)
+		goto fail;
+
+	if (ftruncate(tdb->fd, 0) == -1)
+		goto fail;
+
+	/* This creates an endian-converted header, as if read from disk */
+	CONVERT(*newdb);
+	memcpy(&tdb->header, newdb, sizeof(tdb->header));
+	/* Don't endian-convert the magic food! */
+	memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
+	if (write(tdb->fd, newdb, size) != size) {
+		ret = -1;
+	} else {
+		ret = 0;
+	}
+
+  fail:
+	SAFE_FREE(newdb);
+	return ret;
+}
+
+
+
+static int tdb_already_open(dev_t device,
+			    ino_t ino)
+{
+	struct tdb_context *i;
+	
+	for (i = tdbs; i; i = i->next) {
+		if (i->device == device && i->inode == ino) {
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
+/* open the database, creating it if necessary 
+
+   The open_flags and mode are passed straight to the open call on the
+   database file. A flags value of O_WRONLY is invalid. The hash size
+   is advisory, use zero for a default value.
+
+   Return is NULL on error, in which case errno is also set.  Don't 
+   try to call tdb_error or tdb_errname, just do strerror(errno).
+
+   @param name may be NULL for internal databases. */
+struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
+		      int open_flags, mode_t mode)
+{
+	return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
+}
+
+/* a default logging function */
+static void null_log_fn(struct tdb_context *tdb, int level, const char *fmt, ...)
+{
+}
+
+
+struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+			 int open_flags, mode_t mode,
+			 tdb_log_func log_fn,
+			 tdb_hash_func hash_fn)
+{
+	struct tdb_context *tdb;
+	struct stat st;
+	int rev = 0, locked = 0;
+	unsigned char *vp;
+	u32 vertest;
+
+	if (!(tdb = calloc(1, sizeof *tdb))) {
+		/* Can't log this */
+		errno = ENOMEM;
+		goto fail;
+	}
+	tdb_io_init(tdb);
+	tdb->fd = -1;
+	tdb->name = NULL;
+	tdb->map_ptr = NULL;
+	tdb->flags = tdb_flags;
+	tdb->open_flags = open_flags;
+	tdb->log_fn = log_fn?log_fn:null_log_fn;
+	tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
+
+	/* cache the page size */
+	tdb->page_size = getpagesize();
+	if (tdb->page_size <= 0) {
+		tdb->page_size = 0x2000;
+	}
+
+	if ((open_flags & O_ACCMODE) == O_WRONLY) {
+		TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
+			 name));
+		errno = EINVAL;
+		goto fail;
+	}
+	
+	if (hash_size == 0)
+		hash_size = DEFAULT_HASH_SIZE;
+	if ((open_flags & O_ACCMODE) == O_RDONLY) {
+		tdb->read_only = 1;
+		/* read only databases don't do locking or clear if first */
+		tdb->flags |= TDB_NOLOCK;
+		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+	}
+
+	/* internal databases don't mmap or lock, and start off cleared */
+	if (tdb->flags & TDB_INTERNAL) {
+		tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
+		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+		if (tdb_new_database(tdb, hash_size) != 0) {
+			TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!"));
+			goto fail;
+		}
+		goto internal;
+	}
+
+	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
+		TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
+			 name, strerror(errno)));
+		goto fail;	/* errno set by open(2) */
+	}
+
+	/* ensure there is only one process initialising at once */
+	if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
+			 name, strerror(errno)));
+		goto fail;	/* errno set by tdb_brlock */
+	}
+
+	/* we need to zero database if we are the only one with it open */
+	if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
+	    (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))) {
+		open_flags |= O_CREAT;
+		if (ftruncate(tdb->fd, 0) == -1) {
+			TDB_LOG((tdb, 0, "tdb_open_ex: "
+				 "failed to truncate %s: %s\n",
+				 name, strerror(errno)));
+			goto fail; /* errno set by ftruncate */
+		}
+	}
+
+	if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
+	    || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
+	    || (tdb->header.version != TDB_VERSION
+		&& !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
+		/* its not a valid database - possibly initialise it */
+		if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
+			errno = EIO; /* ie bad format or something */
+			goto fail;
+		}
+		rev = (tdb->flags & TDB_CONVERT);
+	}
+	vp = (unsigned char *)&tdb->header.version;
+	vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
+		  (((u32)vp[2]) << 8) | (u32)vp[3];
+	tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
+	if (!rev)
+		tdb->flags &= ~TDB_CONVERT;
+	else {
+		tdb->flags |= TDB_CONVERT;
+		tdb_convert(&tdb->header, sizeof(tdb->header));
+	}
+	if (fstat(tdb->fd, &st) == -1)
+		goto fail;
+
+	if (tdb->header.rwlocks != 0) {
+		TDB_LOG((tdb, 5, "tdb_open_ex: spinlocks no longer supported\n"));
+		goto fail;
+	}
+
+	/* Is it already in the open list?  If so, fail. */
+	if (tdb_already_open(st.st_dev, st.st_ino)) {
+		TDB_LOG((tdb, 2, "tdb_open_ex: "
+			 "%s (%d,%d) is already open in this process\n",
+			 name, (int)st.st_dev, (int)st.st_ino));
+		errno = EBUSY;
+		goto fail;
+	}
+
+	if (!(tdb->name = (char *)strdup(name))) {
+		errno = ENOMEM;
+		goto fail;
+	}
+
+	tdb->map_size = st.st_size;
+	tdb->device = st.st_dev;
+	tdb->inode = st.st_ino;
+	tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
+	if (!tdb->locked) {
+		TDB_LOG((tdb, 2, "tdb_open_ex: "
+			 "failed to allocate lock structure for %s\n",
+			 name));
+		errno = ENOMEM;
+		goto fail;
+	}
+	tdb_mmap(tdb);
+	if (locked) {
+		if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
+			TDB_LOG((tdb, 0, "tdb_open_ex: "
+				 "failed to take ACTIVE_LOCK on %s: %s\n",
+				 name, strerror(errno)));
+			goto fail;
+		}
+
+	}
+
+	/* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
+	   we didn't get the initial exclusive lock as we need to let all other
+	   users know we're using it. */
+
+	if (tdb_flags & TDB_CLEAR_IF_FIRST) {
+		/* leave this lock in place to indicate it's in use */
+		if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
+			goto fail;
+	}
+
+	/* if needed, run recovery */
+	if (tdb_transaction_recover(tdb) == -1) {
+		goto fail;
+	}
+
+ internal:
+	/* Internal (memory-only) databases skip all the code above to
+	 * do with disk files, and resume here by releasing their
+	 * global lock and hooking into the active list. */
+	if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
+		goto fail;
+	tdb->next = tdbs;
+	tdbs = tdb;
+	return tdb;
+
+ fail:
+	{ int save_errno = errno;
+
+	if (!tdb)
+		return NULL;
+	
+	if (tdb->map_ptr) {
+		if (tdb->flags & TDB_INTERNAL)
+			SAFE_FREE(tdb->map_ptr);
+		else
+			tdb_munmap(tdb);
+	}
+	SAFE_FREE(tdb->name);
+	if (tdb->fd != -1)
+		if (close(tdb->fd) != 0)
+			TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
+	SAFE_FREE(tdb->locked);
+	SAFE_FREE(tdb);
+	errno = save_errno;
+	return NULL;
+	}
+}
+
+/**
+ * Close a database.
+ *
+ * @returns -1 for error; 0 for success.
+ **/
+int tdb_close(struct tdb_context *tdb)
+{
+	struct tdb_context **i;
+	int ret = 0;
+
+	if (tdb->transaction) {
+		tdb_transaction_cancel(tdb);
+	}
+
+	if (tdb->map_ptr) {
+		if (tdb->flags & TDB_INTERNAL)
+			SAFE_FREE(tdb->map_ptr);
+		else
+			tdb_munmap(tdb);
+	}
+	SAFE_FREE(tdb->name);
+	if (tdb->fd != -1)
+		ret = close(tdb->fd);
+	SAFE_FREE(tdb->locked);
+
+	/* Remove from contexts list */
+	for (i = &tdbs; *i; i = &(*i)->next) {
+		if (*i == tdb) {
+			*i = tdb->next;
+			break;
+		}
+	}
+
+	memset(tdb, 0, sizeof(*tdb));
+	SAFE_FREE(tdb);
+
+	return ret;
+}
+
+/* register a loging function */
+void tdb_logging_function(struct tdb_context *tdb, void (*fn)(struct tdb_context *, int , const char *, ...))
+{
+	tdb->log_fn = fn?fn:null_log_fn;
+}
+
+
+/* reopen a tdb - this can be used after a fork to ensure that we have an independent
+   seek pointer from our parent and to re-establish locks */
+int tdb_reopen(struct tdb_context *tdb)
+{
+	struct stat st;
+
+	if (tdb->flags & TDB_INTERNAL) {
+		return 0; /* Nothing to do. */
+	}
+
+	if (tdb->num_locks != 0) {
+		TDB_LOG((tdb, 0, "tdb_reopen: reopen not allowed with locks held\n"));
+		goto fail;
+	}
+
+	if (tdb->transaction != 0) {
+		TDB_LOG((tdb, 0, "tdb_reopen: reopen not allowed inside a transaction\n"));
+		goto fail;
+	}
+
+	if (tdb_munmap(tdb) != 0) {
+		TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	if (close(tdb->fd) != 0)
+		TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
+	tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
+	if (tdb->fd == -1) {
+		TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	if ((tdb->flags & TDB_CLEAR_IF_FIRST) && 
+	    (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)) {
+		TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
+		goto fail;
+	}
+	if (fstat(tdb->fd, &st) != 0) {
+		TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
+		TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
+		goto fail;
+	}
+	tdb_mmap(tdb);
+
+	return 0;
+
+fail:
+	tdb_close(tdb);
+	return -1;
+}
+
+/* reopen all tdb's */
+int tdb_reopen_all(int parent_longlived)
+{
+	struct tdb_context *tdb;
+
+	for (tdb=tdbs; tdb; tdb = tdb->next) {
+		/*
+		 * If the parent is longlived (ie. a
+		 * parent daemon architecture), we know
+		 * it will keep it's active lock on a
+		 * tdb opened with CLEAR_IF_FIRST. Thus
+		 * for child processes we don't have to
+		 * add an active lock. This is essential
+		 * to improve performance on systems that
+		 * keep POSIX locks as a non-scalable data
+		 * structure in the kernel.
+		 */
+		if (parent_longlived) {
+			/* Ensure no clear-if-first. */
+			tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+		}
+
+		if (tdb_reopen(tdb) != 0)
+			return -1;
+	}
+
+	return 0;
+}
diff --git a/source3/tdb/spinlock.c b/source3/tdb/spinlock.c
deleted file mode 100644
index e42a6901c7..0000000000
--- a/source3/tdb/spinlock.c
+++ /dev/null
@@ -1,472 +0,0 @@
-/* 
-   Unix SMB/CIFS implementation.
-
-   trivial database library 
-
-   Copyright (C) Anton Blanchard                   2001
-   
-     ** NOTE! The following LGPL license applies to the tdb
-     ** library. This does NOT imply that all of Samba is released
-     ** under the LGPL
-   
-   This library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2 of the License, or (at your option) any later version.
-
-   This library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-   
-   You should have received a copy of the GNU Lesser General Public
-   License along with this library; if not, write to the Free Software
-   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
-*/
-#if HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#ifdef STANDALONE
-#include <stdlib.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <string.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <sys/stat.h>
-#include <time.h>
-#include <signal.h>
-#include "tdb.h"
-#include "spinlock.h"
-
-#define DEBUG
-#else
-#include "includes.h"
-#endif
-
-#ifdef USE_SPINLOCKS
-
-/*
- * ARCH SPECIFIC
- */
-
-#if defined(SPARC_SPINLOCKS)
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
-	unsigned int result;
-
-	asm volatile("ldstub    [%1], %0"
-		: "=r" (result)
-		: "r" (lock)
-		: "memory");
-
-	return (result == 0) ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
-	asm volatile("":::"memory");
-	*lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
-	*lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
-	return (*lock != 0);
-}
-
-#elif defined(POWERPC_SPINLOCKS) 
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
-	unsigned int result;
-
-	__asm__ __volatile__(
-"1:	lwarx		%0,0,%1\n\
-	cmpwi		0,%0,0\n\
-	li		%0,0\n\
-	bne-		2f\n\
-	li		%0,1\n\
-	stwcx.		%0,0,%1\n\
-	bne-		1b\n\
-	isync\n\
-2:"	: "=&r"(result)
-	: "r"(lock)
-	: "cr0", "memory");
-
-	return (result == 1) ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
-	asm volatile("eieio":::"memory");
-	*lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
-	*lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
-	return (*lock != 0);
-}
-
-#elif defined(INTEL_SPINLOCKS) 
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
-	int oldval;
-
-	asm volatile("xchgl %0,%1"
-		: "=r" (oldval), "=m" (*lock)
-		: "0" (0)
-		: "memory");
-
-	return oldval > 0 ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
-	asm volatile("":::"memory");
-	*lock = 1;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
-	*lock = 1;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
-	return (*lock != 1);
-}
-
-#elif defined(MIPS_SPINLOCKS) && defined(sgi) && (_COMPILER_VERSION >= 730)
-
-/* Implement spinlocks on IRIX using the MIPSPro atomic fetch operations. See
- * sync(3) for the details of the intrinsic operations.
- *
- * "sgi" and "_COMPILER_VERSION" are always defined by MIPSPro.
- */
-
-#ifdef STANDALONE
-
-/* MIPSPro 7.3 has "__inline" as an extension, but not "inline. */
-#define inline __inline
-
-#endif /* STANDALONE */
-
-/* Returns 0 if the lock is acquired, EBUSY otherwise. */
-static inline int __spin_trylock(spinlock_t *lock)
-{
-        unsigned int val;
-        val = __lock_test_and_set(lock, 1);
-        return val == 0 ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
-        __lock_release(lock);
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
-        __lock_release(lock);
-}
-
-/* Returns 1 if the lock is held, 0 otherwise. */
-static inline int __spin_is_locked(spinlock_t *lock)
-{
-        unsigned int val;
-        val = __add_and_fetch(lock, 0);
-	return val;
-}
-
-#elif defined(MIPS_SPINLOCKS) 
-
-static inline unsigned int load_linked(unsigned long addr)
-{
-	unsigned int res;
-
-	__asm__ __volatile__("ll\t%0,(%1)"
-		: "=r" (res)
-		: "r" (addr));
-
-	return res;
-}
-
-static inline unsigned int store_conditional(unsigned long addr, unsigned int value)
-{
-	unsigned int res;
-
-	__asm__ __volatile__("sc\t%0,(%2)"
-		: "=r" (res)
-		: "0" (value), "r" (addr));
-	return res;
-}
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
-	unsigned int mw;
-
-	do {
-		mw = load_linked(lock);
-		if (mw) 
-			return EBUSY;
-	} while (!store_conditional(lock, 1));
-
-	asm volatile("":::"memory");
-
-	return 0;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
-	asm volatile("":::"memory");
-	*lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
-	*lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
-	return (*lock != 0);
-}
-
-#else
-#error Need to implement spinlock code in spinlock.c
-#endif
-
-/*
- * OS SPECIFIC
- */
-
-static void yield_cpu(void)
-{
-	struct timespec tm;
-
-#ifdef USE_SCHED_YIELD
-	sched_yield();
-#else
-	/* Linux will busy loop for delays < 2ms on real time tasks */
-	tm.tv_sec = 0;
-	tm.tv_nsec = 2000000L + 1;
-	nanosleep(&tm, NULL);
-#endif
-}
-
-/*
- * GENERIC
- */
-
-static int smp_machine = 0;
-
-static inline void __spin_lock(spinlock_t *lock)
-{
-	int ntries = 0;
-
-	while(__spin_trylock(lock)) {
-		while(__spin_is_locked(lock)) {
-			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
-				continue;
-			yield_cpu();
-		}
-	}
-}
-
-static void __read_lock(tdb_rwlock_t *rwlock)
-{
-	int ntries = 0;
-
-	while(1) {
-		__spin_lock(&rwlock->lock);
-
-		if (!(rwlock->count & RWLOCK_BIAS)) {
-			rwlock->count++;
-			__spin_unlock(&rwlock->lock);
-			return;
-		}
-	
-		__spin_unlock(&rwlock->lock);
-
-		while(rwlock->count & RWLOCK_BIAS) {
-			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
-				continue;
-			yield_cpu();
-		}
-	}
-}
-
-static void __write_lock(tdb_rwlock_t *rwlock)
-{
-	int ntries = 0;
-
-	while(1) {
-		__spin_lock(&rwlock->lock);
-
-		if (rwlock->count == 0) {
-			rwlock->count |= RWLOCK_BIAS;
-			__spin_unlock(&rwlock->lock);
-			return;
-		}
-
-		__spin_unlock(&rwlock->lock);
-
-		while(rwlock->count != 0) {
-			if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
-				continue;
-			yield_cpu();
-		}
-	}
-}
-
-static void __write_unlock(tdb_rwlock_t *rwlock)
-{
-	__spin_lock(&rwlock->lock);
-
-#ifdef DEBUG
-	if (!(rwlock->count & RWLOCK_BIAS))
-		fprintf(stderr, "bug: write_unlock\n");
-#endif
-
-	rwlock->count &= ~RWLOCK_BIAS;
-	__spin_unlock(&rwlock->lock);
-}
-
-static void __read_unlock(tdb_rwlock_t *rwlock)
-{
-	__spin_lock(&rwlock->lock);
-
-#ifdef DEBUG
-	if (!rwlock->count)
-		fprintf(stderr, "bug: read_unlock\n");
-
-	if (rwlock->count & RWLOCK_BIAS)
-		fprintf(stderr, "bug: read_unlock\n");
-#endif
-
-	rwlock->count--;
-	__spin_unlock(&rwlock->lock);
-}
-
-/* TDB SPECIFIC */
-
-/* lock a list in the database. list -1 is the alloc list */
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type)
-{
-	tdb_rwlock_t *rwlocks;
-
-	if (!tdb->map_ptr) return -1;
-	rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
-
-	switch(rw_type) {
-	case F_RDLCK:
-		__read_lock(&rwlocks[list+1]);
-		break;
-
-	case F_WRLCK:
-		__write_lock(&rwlocks[list+1]);
-		break;
-
-	default:
-		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
-	}
-	return 0;
-}
-
-/* unlock the database. */
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type)
-{
-	tdb_rwlock_t *rwlocks;
-
-	if (!tdb->map_ptr) return -1;
-	rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
-
-	switch(rw_type) {
-	case F_RDLCK:
-		__read_unlock(&rwlocks[list+1]);
-		break;
-
-	case F_WRLCK:
-		__write_unlock(&rwlocks[list+1]);
-		break;
-
-	default:
-		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
-	}
-
-	return 0;
-}
-
-int tdb_create_rwlocks(int fd, unsigned int hash_size)
-{
-	unsigned size, i;
-	tdb_rwlock_t *rwlocks;
-
-	size = TDB_SPINLOCK_SIZE(hash_size);
-	rwlocks = malloc(size);
-	if (!rwlocks)
-		return -1;
-
-	for(i = 0; i < hash_size+1; i++) {
-		__spin_lock_init(&rwlocks[i].lock);
-		rwlocks[i].count = 0;
-	}
-
-	/* Write it out (appending to end) */
-	if (write(fd, rwlocks, size) != size) {
-		free(rwlocks);
-		return -1;
-	}
-	smp_machine = this_is_smp();
-	free(rwlocks);
-	return 0;
-}
-
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
-{
-	tdb_rwlock_t *rwlocks;
-	unsigned i;
-
-	if (tdb->header.rwlocks == 0) return 0;
-	if (!tdb->map_ptr) return -1;
-
-	/* We're mmapped here */
-	rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
-	for(i = 0; i < tdb->header.hash_size+1; i++) {
-		__spin_lock_init(&rwlocks[i].lock);
-		rwlocks[i].count = 0;
-	}
-	return 0;
-}
-#else
-int tdb_create_rwlocks(int fd, unsigned int hash_size) { return 0; }
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
-
-/* Non-spinlock version: remove spinlock pointer */
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
-{
-	tdb_off off = (tdb_off)((char *)&tdb->header.rwlocks
-				- (char *)&tdb->header);
-
-	tdb->header.rwlocks = 0;
-	if (lseek(tdb->fd, off, SEEK_SET) != off
-	    || write(tdb->fd, (void *)&tdb->header.rwlocks,
-		     sizeof(tdb->header.rwlocks)) 
-	    != sizeof(tdb->header.rwlocks))
-		return -1;
-	return 0;
-}
-#endif
diff --git a/source3/tdb/spinlock.h b/source3/tdb/spinlock.h
deleted file mode 100644
index 967fe37457..0000000000
--- a/source3/tdb/spinlock.h
+++ /dev/null
@@ -1,59 +0,0 @@
-#ifndef __SPINLOCK_H__
-#define __SPINLOCK_H__
-
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#include "tdb.h"
-
-#ifdef USE_SPINLOCKS
-
-#define RWLOCK_BIAS 0x1000UL
-
-/* OS SPECIFIC */
-#define MAX_BUSY_LOOPS 1000
-#undef USE_SCHED_YIELD
-
-/* ARCH SPECIFIC */
-/* We should make sure these are padded to a cache line */
-#if defined(SPARC_SPINLOCKS)
-typedef volatile char spinlock_t;
-#elif defined(POWERPC_SPINLOCKS)
-typedef volatile unsigned long spinlock_t;
-#elif defined(INTEL_SPINLOCKS)
-typedef volatile int spinlock_t;
-#elif defined(MIPS_SPINLOCKS)
-typedef volatile unsigned long spinlock_t;
-#else
-#error Need to implement spinlock code in spinlock.h
-#endif
-
-typedef struct {
-	spinlock_t lock;
-	volatile int count;
-} tdb_rwlock_t;
-
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_create_rwlocks(int fd, unsigned int hash_size);
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb);
-
-#define TDB_SPINLOCK_SIZE(hash_size) (((hash_size) + 1) * sizeof(tdb_rwlock_t))
-
-#else /* !USE_SPINLOCKS */
-#if 0
-#define tdb_create_rwlocks(fd, hash_size) 0
-#define tdb_spinlock(tdb, list, rw_type) (-1)
-#define tdb_spinunlock(tdb, list, rw_type) (-1)
-#else
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_create_rwlocks(int fd, unsigned int hash_size);
-#endif
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb);
-#define TDB_SPINLOCK_SIZE(hash_size) 0
-
-#endif
-
-#endif
diff --git a/source3/tdb/tdb.c b/source3/tdb/tdb.c
index ad73a1d9aa..1fae6b6de1 100644
--- a/source3/tdb/tdb.c
+++ b/source3/tdb/tdb.c
@@ -3,7 +3,7 @@
 
    trivial database library
 
-   Copyright (C) Andrew Tridgell              1999-2004
+   Copyright (C) Andrew Tridgell              1999-2005
    Copyright (C) Paul `Rusty' Russell		   2000
    Copyright (C) Jeremy Allison			   2000-2003
    
@@ -20,1004 +20,57 @@
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Lesser General Public License for more details.
-   
+
    You should have received a copy of the GNU Lesser General Public
    License along with this library; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
 
+#include "tdb_private.h"
 
-/* NOTE: If you use tdbs under valgrind, and in particular if you run
- * tdbtorture, you may get spurious "uninitialized value" warnings.  I
- * think this is because valgrind doesn't understand that the mmap'd
- * area may be written to by other processes.  Memory can, from the
- * point of view of the grinded process, spontaneously become
- * initialized.
- *
- * I can think of a few solutions.  [mbp 20030311]
- *
- * 1 - Write suppressions for Valgrind so that it doesn't complain
- * about this.  Probably the most reasonable but people need to
- * remember to use them.
- *
- * 2 - Use IO not mmap when running under valgrind.  Not so nice.
- *
- * 3 - Use the special valgrind macros to mark memory as valid at the
- * right time.  Probably too hard -- the process just doesn't know.
- */ 
-
-#ifdef STANDALONE
-#if HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <string.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <signal.h>
-#include "tdb.h"
-#include "spinlock.h"
-#else
-#include "includes.h"
-
-#if defined(PARANOID_MALLOC_CHECKER)
-#ifdef malloc
-#undef malloc
-#endif
-
-#ifdef realloc
-#undef realloc
-#endif
-
-#ifdef calloc
-#undef calloc
-#endif
-
-#ifdef strdup
-#undef strdup
-#endif
-
-#ifdef strndup
-#undef strndup
-#endif
-
-#endif
-
-#endif
-
-#define TDB_MAGIC_FOOD "TDB file\n"
-#define TDB_VERSION (0x26011967 + 6)
-#define TDB_MAGIC (0x26011999U)
-#define TDB_FREE_MAGIC (~TDB_MAGIC)
-#define TDB_DEAD_MAGIC (0xFEE1DEAD)
-#define TDB_ALIGNMENT 4
-#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
-#define DEFAULT_HASH_SIZE 131
-#define TDB_PAGE_SIZE 0x2000
-#define FREELIST_TOP (sizeof(struct tdb_header))
-#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
-#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
-#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
-#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
-#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
-#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + TDB_SPINLOCK_SIZE(hash_size))
-
-
-/* NB assumes there is a local variable called "tdb" that is the
- * current context, also takes doubly-parenthesized print-style
- * argument. */
-#define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0)
-
-/* lock offsets */
-#define GLOBAL_LOCK 0
-#define ACTIVE_LOCK 4
-
-#ifndef MAP_FILE
-#define MAP_FILE 0
-#endif
-
-#ifndef MAP_FAILED
-#define MAP_FAILED ((void *)-1)
-#endif
-
-/* free memory if the pointer is valid and zero the pointer */
-#ifndef SAFE_FREE
-#define SAFE_FREE(x) do { if ((x) != NULL) {free((x)); (x)=NULL;} } while(0)
-#endif
-
-#define BUCKET(hash) ((hash) % tdb->header.hash_size)
 TDB_DATA tdb_null;
 
-/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
-static TDB_CONTEXT *tdbs = NULL;
-
-static int tdb_munmap(TDB_CONTEXT *tdb)
+/*
+  increment the tdb sequence number if the tdb has been opened using
+  the TDB_SEQNUM flag
+*/
+static void tdb_increment_seqnum(struct tdb_context *tdb)
 {
-	if (tdb->flags & TDB_INTERNAL)
-		return 0;
-
-#ifdef HAVE_MMAP
-	if (tdb->map_ptr) {
-		int ret = munmap(tdb->map_ptr, tdb->map_size);
-		if (ret != 0)
-			return ret;
+	tdb_off_t seqnum=0;
+	
+	if (!(tdb->flags & TDB_SEQNUM)) {
+		return;
 	}
-#endif
-	tdb->map_ptr = NULL;
-	return 0;
-}
 
-static void tdb_mmap(TDB_CONTEXT *tdb)
-{
-	if (tdb->flags & TDB_INTERNAL)
+	if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1) != 0) {
 		return;
-
-#ifdef HAVE_MMAP
-	if (!(tdb->flags & TDB_NOMMAP)) {
-		tdb->map_ptr = mmap(NULL, tdb->map_size, 
-				    PROT_READ|(tdb->read_only? 0:PROT_WRITE), 
-				    MAP_SHARED|MAP_FILE, tdb->fd, 0);
-
-		/*
-		 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
-		 */
-
-		if (tdb->map_ptr == MAP_FAILED) {
-			tdb->map_ptr = NULL;
-			TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n", 
-				 tdb->map_size, strerror(errno)));
-		}
-	} else {
-		tdb->map_ptr = NULL;
 	}
-#else
-	tdb->map_ptr = NULL;
-#endif
-}
 
-/* Endian conversion: we only ever deal with 4 byte quantities */
-static void *convert(void *buf, u32 size)
-{
-	u32 i, *p = buf;
-	for (i = 0; i < size / 4; i++)
-		p[i] = TDB_BYTEREV(p[i]);
-	return buf;
-}
-#define DOCONV() (tdb->flags & TDB_CONVERT)
-#define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
-
-/* the body of the database is made of one list_struct for the free space
-   plus a separate data list for each hash value */
-struct list_struct {
-	tdb_off next; /* offset of the next record in the list */
-	tdb_len rec_len; /* total byte length of record */
-	tdb_len key_len; /* byte length of key */
-	tdb_len data_len; /* byte length of data */
-	u32 full_hash; /* the full 32 bit hash of the key */
-	u32 magic;   /* try to catch errors */
-	/* the following union is implied:
-		union {
-			char record[rec_len];
-			struct {
-				char key[key_len];
-				char data[data_len];
-			}
-			u32 totalsize; (tailer)
-		}
+	/* we ignore errors from this, as we have no sane way of
+	   dealing with them.
 	*/
-};
-
-/***************************************************************
- Allow a caller to set a "alarm" flag that tdb can check to abort
- a blocking lock on SIGALRM.
-***************************************************************/
-
-static sig_atomic_t *palarm_fired;
-
-void tdb_set_lock_alarm(sig_atomic_t *palarm)
-{
-	palarm_fired = palarm;
-}
-
-/* a byte range locking function - return 0 on success
-   this functions locks/unlocks 1 byte at the specified offset.
-
-   On error, errno is also set so that errors are passed back properly
-   through tdb_open(). */
-static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset, 
-		      int rw_type, int lck_type, int probe)
-{
-	struct flock fl;
-	int ret;
-
-	if (tdb->flags & TDB_NOLOCK)
-		return 0;
-	if ((rw_type == F_WRLCK) && (tdb->read_only)) {
-		errno = EACCES;
-		return -1;
-	}
-
-	fl.l_type = rw_type;
-	fl.l_whence = SEEK_SET;
-	fl.l_start = offset;
-	fl.l_len = 1;
-	fl.l_pid = 0;
-
-	do {
-		ret = fcntl(tdb->fd,lck_type,&fl);
-		if (ret == -1 && errno == EINTR && palarm_fired && *palarm_fired)
-			break;
-	} while (ret == -1 && errno == EINTR);
-
-	if (ret == -1) {
-		if (!probe && lck_type != F_SETLK) {
-			/* Ensure error code is set for log fun to examine. */
-			if (errno == EINTR && palarm_fired && *palarm_fired)
-				tdb->ecode = TDB_ERR_LOCK_TIMEOUT;
-			else
-				tdb->ecode = TDB_ERR_LOCK;
-			TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n", 
-				 tdb->fd, offset, rw_type, lck_type));
-		}
-		/* Was it an alarm timeout ? */
-		if (errno == EINTR && palarm_fired && *palarm_fired) {
-			TDB_LOG((tdb, 5, "tdb_brlock timed out (fd=%d) at offset %d rw_type=%d lck_type=%d\n", 
-				 tdb->fd, offset, rw_type, lck_type));
-			return TDB_ERRCODE(TDB_ERR_LOCK_TIMEOUT, -1);
-		}
-		/* Otherwise - generic lock error. errno set by fcntl.
-		 * EAGAIN is an expected return from non-blocking
-		 * locks. */
-		if (errno != EAGAIN) {
-			TDB_LOG((tdb, 5, "tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d: %s\n", 
-				 tdb->fd, offset, rw_type, lck_type, 
-				 strerror(errno)));
-		}
-		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
-	}
-	return 0;
-}
-
-/* lock a list in the database. list -1 is the alloc list */
-static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
-{
-	if (list < -1 || list >= (int)tdb->header.hash_size) {
-		TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n", 
-			   list, ltype));
-		return -1;
-	}
-	if (tdb->flags & TDB_NOLOCK)
-		return 0;
-
-	/* Since fcntl locks don't nest, we do a lock for the first one,
-	   and simply bump the count for future ones */
-	if (tdb->locked[list+1].count == 0) {
-		if (!tdb->read_only && tdb->header.rwlocks) {
-			if (tdb_spinlock(tdb, list, ltype)) {
-				TDB_LOG((tdb, 0, "tdb_lock spinlock failed on list %d ltype=%d\n", 
-					   list, ltype));
-				return -1;
-			}
-		} else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
-			TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n", 
-					   list, ltype, strerror(errno)));
-			return -1;
-		}
-		tdb->locked[list+1].ltype = ltype;
-	}
-	tdb->locked[list+1].count++;
-	return 0;
-}
-
-/* unlock the database: returns void because it's too late for errors. */
-	/* changed to return int it may be interesting to know there
-	   has been an error  --simo */
-static int tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype)
-{
-	int ret = -1;
-
-	if (tdb->flags & TDB_NOLOCK)
-		return 0;
-
-	/* Sanity checks */
-	if (list < -1 || list >= (int)tdb->header.hash_size) {
-		TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
-		return ret;
-	}
-
-	if (tdb->locked[list+1].count==0) {
-		TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
-		return ret;
-	}
-
-	if (tdb->locked[list+1].count == 1) {
-		/* Down to last nested lock: unlock underneath */
-		if (!tdb->read_only && tdb->header.rwlocks) {
-			ret = tdb_spinunlock(tdb, list, ltype);
-		} else {
-			ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
-		}
-	} else {
-		ret = 0;
-	}
-	tdb->locked[list+1].count--;
-
-	if (ret)
-		TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n")); 
-	return ret;
-}
-
-/* check for an out of bounds access - if it is out of bounds then
-   see if the database has been expanded by someone else and expand
-   if necessary 
-   note that "len" is the minimum length needed for the db
-*/
-static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
-{
-	struct stat st;
-	if (len <= tdb->map_size)
-		return 0;
-	if (tdb->flags & TDB_INTERNAL) {
-		if (!probe) {
-			/* Ensure ecode is set for log fn. */
-			tdb->ecode = TDB_ERR_IO;
-			TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n",
-				 (int)len, (int)tdb->map_size));
-		}
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-	}
-
-	if (fstat(tdb->fd, &st) == -1)
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-
-	if (st.st_size < (size_t)len) {
-		if (!probe) {
-			/* Ensure ecode is set for log fn. */
-			tdb->ecode = TDB_ERR_IO;
-			TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
-				 (int)len, (int)st.st_size));
-		}
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-	}
-
-	/* Unmap, update size, remap */
-	if (tdb_munmap(tdb) == -1)
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-	tdb->map_size = st.st_size;
-	tdb_mmap(tdb);
-	return 0;
-}
-
-/* write a lump of data at a specified offset */
-static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
-{
-	if (tdb_oob(tdb, off + len, 0) != 0)
-		return -1;
-
-	if (tdb->map_ptr)
-		memcpy(off + (char *)tdb->map_ptr, buf, len);
-#ifdef HAVE_PWRITE
-	else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
-#else
-	else if (lseek(tdb->fd, off, SEEK_SET) != off
-		 || write(tdb->fd, buf, len) != (ssize_t)len) {
-#endif
-		/* Ensure ecode is set for log fn. */
-		tdb->ecode = TDB_ERR_IO;
-		TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
-			   off, len, strerror(errno)));
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-	}
-	return 0;
-}
-
-/* read a lump of data at a specified offset, maybe convert */
-static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
-{
-	if (tdb_oob(tdb, off + len, 0) != 0)
-		return -1;
-
-	if (tdb->map_ptr)
-		memcpy(buf, off + (char *)tdb->map_ptr, len);
-#ifdef HAVE_PREAD
-	else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) {
-#else
-	else if (lseek(tdb->fd, off, SEEK_SET) != off
-		 || read(tdb->fd, buf, len) != (ssize_t)len) {
-#endif
-		/* Ensure ecode is set for log fn. */
-		tdb->ecode = TDB_ERR_IO;
-		TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
-			   off, len, strerror(errno)));
-		return TDB_ERRCODE(TDB_ERR_IO, -1);
-	}
-	if (cv)
-		convert(buf, len);
-	return 0;
-}
-
-/* read a lump of data, allocating the space for it */
-static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
-{
-	char *buf;
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	seqnum++;
+	tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
 
-	if (!(buf = malloc(len))) {
-		/* Ensure ecode is set for log fn. */
-		tdb->ecode = TDB_ERR_OOM;
-		TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
-			   len, strerror(errno)));
-		return TDB_ERRCODE(TDB_ERR_OOM, buf);
-	}
-	if (tdb_read(tdb, offset, buf, len, 0) == -1) {
-		SAFE_FREE(buf);
-		return NULL;
-	}
-	return buf;
-}
-
-/* read/write a tdb_off */
-static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
-{
-	return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
-}
-static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
-{
-	tdb_off off = *d;
-	return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
+	tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1);
 }
 
-/* read/write a record */
-static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
-	if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
-		return -1;
-	if (TDB_BAD_MAGIC(rec)) {
-		/* Ensure ecode is set for log fn. */
-		tdb->ecode = TDB_ERR_CORRUPT;
-		TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
-		return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
-	}
-	return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
-}
-static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
-	struct list_struct r = *rec;
-	return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
-}
-
-/* read a freelist record and check for simple errors */
-static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec)
-{
-	if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
-		return -1;
-
-	if (rec->magic == TDB_MAGIC) {
-		/* this happens when a app is showdown while deleting a record - we should
-		   not completely fail when this happens */
-		TDB_LOG((tdb, 0,"rec_free_read non-free magic 0x%x at offset=%d - fixing\n", 
-			 rec->magic, off));
-		rec->magic = TDB_FREE_MAGIC;
-		if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
-			return -1;
-	}
-
-	if (rec->magic != TDB_FREE_MAGIC) {
-		/* Ensure ecode is set for log fn. */
-		tdb->ecode = TDB_ERR_CORRUPT;
-		TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n", 
-			   rec->magic, off));
-		return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
-	}
-	if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
-		return -1;
-	return 0;
-}
-
-/* update a record tailer (must hold allocation lock) */
-static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
-			 const struct list_struct *rec)
-{
-	tdb_off totalsize;
-
-	/* Offset of tailer from record header */
-	totalsize = sizeof(*rec) + rec->rec_len;
-	return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
-			 &totalsize);
-}
-
-static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
-{
-	struct list_struct rec;
-	tdb_off tailer_ofs, tailer;
-
-	if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
-		printf("ERROR: failed to read record at %u\n", offset);
-		return 0;
-	}
-
-	printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
-	       offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
-
-	tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
-	if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
-		printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
-		return rec.next;
-	}
-
-	if (tailer != rec.rec_len + sizeof(rec)) {
-		printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
-				(unsigned)tailer, (unsigned)(rec.rec_len + sizeof(rec)));
-	}
-	return rec.next;
-}
-
-static int tdb_dump_chain(TDB_CONTEXT *tdb, int i)
-{
-	tdb_off rec_ptr, top;
-	int hash_length = 0;
-
-	top = TDB_HASH_TOP(i);
-
-	if (tdb_lock(tdb, i, F_WRLCK) != 0)
-		return -1;
-
-	if (ofs_read(tdb, top, &rec_ptr) == -1)
-		return tdb_unlock(tdb, i, F_WRLCK);
-
-	if (rec_ptr)
-		printf("hash=%d\n", i);
-
-	while (rec_ptr) {
-		rec_ptr = tdb_dump_record(tdb, rec_ptr);
-		hash_length += 1;
-	}
-
-	printf("chain %d length %d\n", i, hash_length);
-
-	return tdb_unlock(tdb, i, F_WRLCK);
-}
-
-void tdb_dump_all(TDB_CONTEXT *tdb)
-{
-	int i;
-	for (i=0;i<tdb->header.hash_size;i++) {
-		tdb_dump_chain(tdb, i);
-	}
-	tdb_printfreelist(tdb);
-}
-
-int tdb_printfreelist(TDB_CONTEXT *tdb)
-{
-	int ret;
-	long total_free = 0;
-	tdb_off offset, rec_ptr;
-	struct list_struct rec;
-
-	if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
-		return ret;
-
-	offset = FREELIST_TOP;
-
-	/* read in the freelist top */
-	if (ofs_read(tdb, offset, &rec_ptr) == -1) {
-		tdb_unlock(tdb, -1, F_WRLCK);
-		return 0;
-	}
-
-	printf("freelist top=[0x%08x]\n", rec_ptr );
-	while (rec_ptr) {
-		if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
-			tdb_unlock(tdb, -1, F_WRLCK);
-			return -1;
-		}
-
-		if (rec.magic != TDB_FREE_MAGIC) {
-			printf("bad magic 0x%08x in free list\n", rec.magic);
-			tdb_unlock(tdb, -1, F_WRLCK);
-			return -1;
-		}
-
-		printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec.next, rec.rec_len, rec.rec_len );
-		total_free += rec.rec_len;
-
-		/* move to the next record */
-		rec_ptr = rec.next;
-	}
-	printf("total rec_len = [0x%08x (%d)]\n", (int)total_free, 
-               (int)total_free);
-
-	return tdb_unlock(tdb, -1, F_WRLCK);
-}
-
-/* Remove an element from the freelist.  Must have alloc lock. */
-static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
-{
-	tdb_off last_ptr, i;
-
-	/* read in the freelist top */
-	last_ptr = FREELIST_TOP;
-	while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
-		if (i == off) {
-			/* We've found it! */
-			return ofs_write(tdb, last_ptr, &next);
-		}
-		/* Follow chain (next offset is at start of record) */
-		last_ptr = i;
-	}
-	TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
-	return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
-}
-
-/* Add an element into the freelist. Merge adjacent records if
-   neccessary. */
-static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
-	tdb_off right, left;
-
-	/* Allocation and tailer lock */
-	if (tdb_lock(tdb, -1, F_WRLCK) != 0)
-		return -1;
-
-	/* set an initial tailer, so if we fail we don't leave a bogus record */
-	if (update_tailer(tdb, offset, rec) != 0) {
-		TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
-		goto fail;
-	}
-
-	/* Look right first (I'm an Australian, dammit) */
-	right = offset + sizeof(*rec) + rec->rec_len;
-	if (right + sizeof(*rec) <= tdb->map_size) {
-		struct list_struct r;
-
-		if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
-			TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
-			goto left;
-		}
-
-		/* If it's free, expand to include it. */
-		if (r.magic == TDB_FREE_MAGIC) {
-			if (remove_from_freelist(tdb, right, r.next) == -1) {
-				TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
-				goto left;
-			}
-			rec->rec_len += sizeof(r) + r.rec_len;
-		}
-	}
-
-left:
-	/* Look left */
-	left = offset - sizeof(tdb_off);
-	if (left > TDB_DATA_START(tdb->header.hash_size)) {
-		struct list_struct l;
-		tdb_off leftsize;
-		
-		/* Read in tailer and jump back to header */
-		if (ofs_read(tdb, left, &leftsize) == -1) {
-			TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
-			goto update;
-		}
-		left = offset - leftsize;
-
-		/* Now read in record */
-		if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
-			TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
-			goto update;
-		}
-
-		/* If it's free, expand to include it. */
-		if (l.magic == TDB_FREE_MAGIC) {
-			if (remove_from_freelist(tdb, left, l.next) == -1) {
-				TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
-				goto update;
-			} else {
-				offset = left;
-				rec->rec_len += leftsize;
-			}
-		}
-	}
-
-update:
-	if (update_tailer(tdb, offset, rec) == -1) {
-		TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
-		goto fail;
-	}
-
-	/* Now, prepend to free list */
-	rec->magic = TDB_FREE_MAGIC;
-
-	if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
-	    rec_write(tdb, offset, rec) == -1 ||
-	    ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
-		TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
-		goto fail;
-	}
-
-	/* And we're done. */
-	tdb_unlock(tdb, -1, F_WRLCK);
-	return 0;
-
- fail:
-	tdb_unlock(tdb, -1, F_WRLCK);
-	return -1;
-}
-
-
-/* expand a file.  we prefer to use ftruncate, as that is what posix
-  says to use for mmap expansion */
-static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
-{
-	char buf[1024];
-#if HAVE_FTRUNCATE_EXTEND
-	if (ftruncate(tdb->fd, size+addition) != 0) {
-		TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n", 
-			   size+addition, strerror(errno)));
-		return -1;
-	}
-#else
-	char b = 0;
-
-#ifdef HAVE_PWRITE
-	if (pwrite(tdb->fd,  &b, 1, (size+addition) - 1) != 1) {
-#else
-	if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 || 
-	    write(tdb->fd, &b, 1) != 1) {
-#endif
-		TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n", 
-			   size+addition, strerror(errno)));
-		return -1;
-	}
-#endif
-
-	/* now fill the file with something. This ensures that the file isn't sparse, which would be
-	   very bad if we ran out of disk. This must be done with write, not via mmap */
-	memset(buf, 0x42, sizeof(buf));
-	while (addition) {
-		int n = addition>sizeof(buf)?sizeof(buf):addition;
-#ifdef HAVE_PWRITE
-		int ret = pwrite(tdb->fd, buf, n, size);
-#else
-		int ret;
-		if (lseek(tdb->fd, size, SEEK_SET) != size)
-			return -1;
-		ret = write(tdb->fd, buf, n);
-#endif
-		if (ret != n) {
-			TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n", 
-				   n, strerror(errno)));
-			return -1;
-		}
-		addition -= n;
-		size += n;
-	}
-	return 0;
-}
-
-
-/* expand the database at least size bytes by expanding the underlying
-   file and doing the mmap again if necessary */
-static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
-{
-	struct list_struct rec;
-	tdb_off offset;
-
-	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
-		TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
-		return -1;
-	}
-
-	/* must know about any previous expansions by another process */
-	tdb_oob(tdb, tdb->map_size + 1, 1);
-
-	/* always make room for at least 10 more records, and round
-           the database up to a multiple of TDB_PAGE_SIZE */
-	size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size;
-
-	if (!(tdb->flags & TDB_INTERNAL))
-		tdb_munmap(tdb);
-
-	/*
-	 * We must ensure the file is unmapped before doing this
-	 * to ensure consistency with systems like OpenBSD where
-	 * writes and mmaps are not consistent.
-	 */
-
-	/* expand the file itself */
-	if (!(tdb->flags & TDB_INTERNAL)) {
-		if (expand_file(tdb, tdb->map_size, size) != 0)
-			goto fail;
-	}
-
-	tdb->map_size += size;
-
-	if (tdb->flags & TDB_INTERNAL) {
-		char *new_map_ptr = realloc(tdb->map_ptr, tdb->map_size);
-		if (!new_map_ptr) {
-			tdb->map_size -= size;
-			goto fail;
-		}
-		tdb->map_ptr = new_map_ptr;
-	} else {
-		/*
-		 * We must ensure the file is remapped before adding the space
-		 * to ensure consistency with systems like OpenBSD where
-		 * writes and mmaps are not consistent.
-		 */
-
-		/* We're ok if the mmap fails as we'll fallback to read/write */
-		tdb_mmap(tdb);
-	}
-
-	/* form a new freelist record */
-	memset(&rec,'\0',sizeof(rec));
-	rec.rec_len = size - sizeof(rec);
-
-	/* link it into the free list */
-	offset = tdb->map_size - size;
-	if (tdb_free(tdb, offset, &rec) == -1)
-		goto fail;
-
-	tdb_unlock(tdb, -1, F_WRLCK);
-	return 0;
- fail:
-	tdb_unlock(tdb, -1, F_WRLCK);
-	return -1;
-}
-
-/* allocate some space from the free list. The offset returned points
-   to a unconnected list_struct within the database with room for at
-   least length bytes of total data
-
-   0 is returned if the space could not be allocated
- */
-static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
-			    struct list_struct *rec)
-{
-	tdb_off rec_ptr, last_ptr, newrec_ptr;
-	struct list_struct newrec;
-
-	memset(&newrec, '\0', sizeof(newrec));
-
-	if (tdb_lock(tdb, -1, F_WRLCK) == -1)
-		return 0;
-
-	/* Extra bytes required for tailer */
-	length += sizeof(tdb_off);
-
- again:
-	last_ptr = FREELIST_TOP;
-
-	/* read in the freelist top */
-	if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
-		goto fail;
-
-	/* keep looking until we find a freelist record big enough */
-	while (rec_ptr) {
-		if (rec_free_read(tdb, rec_ptr, rec) == -1)
-			goto fail;
-
-		if (rec->rec_len >= length) {
-			/* found it - now possibly split it up  */
-			if (rec->rec_len > length + MIN_REC_SIZE) {
-				/* Length of left piece */
-				length = TDB_ALIGN(length, TDB_ALIGNMENT);
-
-				/* Right piece to go on free list */
-				newrec.rec_len = rec->rec_len
-					- (sizeof(*rec) + length);
-				newrec_ptr = rec_ptr + sizeof(*rec) + length;
-
-				/* And left record is shortened */
-				rec->rec_len = length;
-			} else
-				newrec_ptr = 0;
-
-			/* Remove allocated record from the free list */
-			if (ofs_write(tdb, last_ptr, &rec->next) == -1)
-				goto fail;
-
-			/* Update header: do this before we drop alloc
-                           lock, otherwise tdb_free() might try to
-                           merge with us, thinking we're free.
-                           (Thanks Jeremy Allison). */
-			rec->magic = TDB_MAGIC;
-			if (rec_write(tdb, rec_ptr, rec) == -1)
-				goto fail;
-
-			/* Did we create new block? */
-			if (newrec_ptr) {
-				/* Update allocated record tailer (we
-                                   shortened it). */
-				if (update_tailer(tdb, rec_ptr, rec) == -1)
-					goto fail;
-
-				/* Free new record */
-				if (tdb_free(tdb, newrec_ptr, &newrec) == -1)
-					goto fail;
-			}
-
-			/* all done - return the new record offset */
-			tdb_unlock(tdb, -1, F_WRLCK);
-			return rec_ptr;
-		}
-		/* move to the next record */
-		last_ptr = rec_ptr;
-		rec_ptr = rec->next;
-	}
-	/* we didn't find enough space. See if we can expand the
-	   database and if we can then try again */
-	if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
-		goto again;
- fail:
-	tdb_unlock(tdb, -1, F_WRLCK);
-	return 0;
-}
-
-/* initialise a new database with a specified hash size */
-static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
-{
-	struct tdb_header *newdb;
-	int size, ret = -1;
-
-	/* We make it up in memory, then write it out if not internal */
-	size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
-	if (!(newdb = calloc(size, 1)))
-		return TDB_ERRCODE(TDB_ERR_OOM, -1);
-
-	/* Fill in the header */
-	newdb->version = TDB_VERSION;
-	newdb->hash_size = hash_size;
-#ifdef USE_SPINLOCKS
-	newdb->rwlocks = size;
-#endif
-	if (tdb->flags & TDB_INTERNAL) {
-		tdb->map_size = size;
-		tdb->map_ptr = (char *)newdb;
-		memcpy(&tdb->header, newdb, sizeof(tdb->header));
-		/* Convert the `ondisk' version if asked. */
-		CONVERT(*newdb);
-		return 0;
-	}
-	if (lseek(tdb->fd, 0, SEEK_SET) == -1)
-		goto fail;
-
-	if (ftruncate(tdb->fd, 0) == -1)
-		goto fail;
-
-	/* This creates an endian-converted header, as if read from disk */
-	CONVERT(*newdb);
-	memcpy(&tdb->header, newdb, sizeof(tdb->header));
-	/* Don't endian-convert the magic food! */
-	memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
-	if (write(tdb->fd, newdb, size) != size)
-		ret = -1;
-	else
-		ret = tdb_create_rwlocks(tdb->fd, hash_size);
-
-  fail:
-	SAFE_FREE(newdb);
-	return ret;
-}
 
 /* Returns 0 on fail.  On success, return offset of record, and fills
    in rec */
-static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
+static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, u32 hash,
 			struct list_struct *r)
 {
-	tdb_off rec_ptr;
+	tdb_off_t rec_ptr;
 	
 	/* read in the hash top */
-	if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
 		return 0;
 
 	/* keep looking until we find the right record */
 	while (rec_ptr) {
-		if (rec_read(tdb, rec_ptr, r) == -1)
+		if (tdb_rec_read(tdb, rec_ptr, r) == -1)
 			return 0;
 
 		if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) {
@@ -1040,8 +93,8 @@ static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
 }
 
 /* As tdb_find, but if you succeed, keep the lock */
-static tdb_off tdb_find_lock_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, int locktype,
-			     struct list_struct *rec)
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash, int locktype,
+			   struct list_struct *rec)
 {
 	u32 rec_ptr;
 
@@ -1052,60 +105,34 @@ static tdb_off tdb_find_lock_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, int
 	return rec_ptr;
 }
 
-enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
-{
-	return tdb->ecode;
-}
-
-static struct tdb_errname {
-	enum TDB_ERROR ecode; const char *estring;
-} emap[] = { {TDB_SUCCESS, "Success"},
-	     {TDB_ERR_CORRUPT, "Corrupt database"},
-	     {TDB_ERR_IO, "IO Error"},
-	     {TDB_ERR_LOCK, "Locking error"},
-	     {TDB_ERR_OOM, "Out of memory"},
-	     {TDB_ERR_EXISTS, "Record exists"},
-	     {TDB_ERR_NOLOCK, "Lock exists on other keys"},
-	     {TDB_ERR_NOEXIST, "Record does not exist"} };
-
-/* Error string for the last tdb error */
-const char *tdb_errorstr(TDB_CONTEXT *tdb)
-{
-	u32 i;
-	for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
-		if (tdb->ecode == emap[i].ecode)
-			return emap[i].estring;
-	return "Invalid error code";
-}
 
 /* update an entry in place - this only works if the new data size
    is <= the old data size and the key exists.
    on failure return -1.
 */
-
-static int tdb_update_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA dbuf)
+static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash, TDB_DATA dbuf)
 {
 	struct list_struct rec;
-	tdb_off rec_ptr;
+	tdb_off_t rec_ptr;
 
 	/* find entry */
 	if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
 		return -1;
 
 	/* must be long enough key, data and tailer */
-	if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
+	if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off_t)) {
 		tdb->ecode = TDB_SUCCESS; /* Not really an error */
 		return -1;
 	}
 
-	if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+	if (tdb->methods->tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
 		      dbuf.dptr, dbuf.dsize) == -1)
 		return -1;
 
 	if (dbuf.dsize != rec.data_len) {
 		/* update size */
 		rec.data_len = dbuf.dsize;
-		return rec_write(tdb, rec_ptr, &rec);
+		return tdb_rec_write(tdb, rec_ptr, &rec);
 	}
  
 	return 0;
@@ -1114,13 +141,12 @@ static int tdb_update_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA db
 /* find an entry in the database given a key */
 /* If an entry doesn't exist tdb_err will be set to
  * TDB_ERR_NOEXIST. If a key has no data attached
- * tdb_err will not be set. Both will return a
- * zero pptr and zero dsize.
+ * then the TDB_DATA will have zero length but
+ * a non-zero pointer
  */
-
-TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
+TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
 {
-	tdb_off rec_ptr;
+	tdb_off_t rec_ptr;
 	struct list_struct rec;
 	TDB_DATA ret;
 	u32 hash;
@@ -1130,11 +156,8 @@ TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
 	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
 		return tdb_null;
 
-	if (rec.data_len)
-		ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
-					  rec.data_len);
-	else
-		ret.dptr = NULL;
+	ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+				  rec.data_len);
 	ret.dsize = rec.data_len;
 	tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
 	return ret;
@@ -1146,7 +169,7 @@ TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
    this doesn't match the conventions in the rest of this module, but is
    compatible with gdbm
 */
-static int tdb_exists_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
+static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash)
 {
 	struct list_struct rec;
 	
@@ -1156,82 +179,39 @@ static int tdb_exists_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
 	return 1;
 }
 
-int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
+int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
 {
 	u32 hash = tdb->hash_fn(&key);
 	return tdb_exists_hash(tdb, key, hash);
 }
 
-/* record lock stops delete underneath */
-static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
-	return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
-}
-/*
-  Write locks override our own fcntl readlocks, so check it here.
-  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
-  an error to fail to get the lock here.
-*/
- 
-static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
-	struct tdb_traverse_lock *i;
-	for (i = &tdb->travlocks; i; i = i->next)
-		if (i->off == off)
-			return -1;
-	return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
-}
-
-/*
-  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
-  an error to fail to get the lock here.
-*/
-
-static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
-	return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
-}
-/* fcntl locks don't stack: avoid unlocking someone else's */
-static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
-	struct tdb_traverse_lock *i;
-	u32 count = 0;
-
-	if (off == 0)
-		return 0;
-	for (i = &tdb->travlocks; i; i = i->next)
-		if (i->off == off)
-			count++;
-	return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
-}
-
 /* actually delete an entry in the database given the offset */
-static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
+int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct list_struct*rec)
 {
-	tdb_off last_ptr, i;
+	tdb_off_t last_ptr, i;
 	struct list_struct lastrec;
 
-	if (tdb->read_only) return -1;
+	if (tdb->read_only || tdb->traverse_read) return -1;
 
-	if (write_lock_record(tdb, rec_ptr) == -1) {
+	if (tdb_write_lock_record(tdb, rec_ptr) == -1) {
 		/* Someone traversing here: mark it as dead */
 		rec->magic = TDB_DEAD_MAGIC;
-		return rec_write(tdb, rec_ptr, rec);
+		return tdb_rec_write(tdb, rec_ptr, rec);
 	}
-	if (write_unlock_record(tdb, rec_ptr) != 0)
+	if (tdb_write_unlock_record(tdb, rec_ptr) != 0)
 		return -1;
 
 	/* find previous record in hash chain */
-	if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
 		return -1;
 	for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
-		if (rec_read(tdb, i, &lastrec) == -1)
+		if (tdb_rec_read(tdb, i, &lastrec) == -1)
 			return -1;
 
 	/* unlink it: next ptr is at start of record. */
 	if (last_ptr == 0)
 		last_ptr = TDB_HASH_TOP(rec->full_hash);
-	if (ofs_write(tdb, last_ptr, &rec->next) == -1)
+	if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1)
 		return -1;
 
 	/* recover the space */
@@ -1240,283 +220,27 @@ static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
 	return 0;
 }
 
-/* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
-static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
-			 struct list_struct *rec)
-{
-	int want_next = (tlock->off != 0);
-
-	/* Lock each chain from the start one. */
-	for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
-
-		/* this is an optimisation for the common case where
-		   the hash chain is empty, which is particularly
-		   common for the use of tdb with ldb, where large
-		   hashes are used. In that case we spend most of our
-		   time in tdb_brlock(), locking empty hash chains.
-
-		   To avoid this, we do an unlocked pre-check to see
-		   if the hash chain is empty before starting to look
-		   inside it. If it is empty then we can avoid that
-		   hash chain. If it isn't empty then we can't believe
-		   the value we get back, as we read it without a
-		   lock, so instead we get the lock and re-fetch the
-		   value below.
-
-		   Notice that not doing this optimisation on the
-		   first hash chain is critical. We must guarantee
-		   that we have done at least one fcntl lock at the
-		   start of a search to guarantee that memory is
-		   coherent on SMP systems. If records are added by
-		   others during the search then thats OK, and we
-		   could possibly miss those with this trick, but we
-		   could miss them anyway without this trick, so the
-		   semantics don't change.
-
-		   With a non-indexed ldb search this trick gains us a
-		   factor of around 80 in speed on a linux 2.6.x
-		   system (testing using ldbtest).
-		 */
-		if (!tlock->off && tlock->hash != 0) {
-			u32 off;
-			if (tdb->map_ptr) {
-				for (;tlock->hash < tdb->header.hash_size;tlock->hash++) {
-					if (0 != *(u32 *)(TDB_HASH_TOP(tlock->hash) + (unsigned char *)tdb->map_ptr)) {
-						break;
-					}
-				}
-				if (tlock->hash == tdb->header.hash_size) {
-					continue;
-				}
-			} else {
-				if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash), &off) == 0 &&
-				    off == 0) {
-					continue;
-				}
-			}
-		}
-
-		if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
-			return -1;
-
-		/* No previous record?  Start at top of chain. */
-		if (!tlock->off) {
-			if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
-				     &tlock->off) == -1)
-				goto fail;
-		} else {
-			/* Otherwise unlock the previous record. */
-			if (unlock_record(tdb, tlock->off) != 0)
-				goto fail;
-		}
-
-		if (want_next) {
-			/* We have offset of old record: grab next */
-			if (rec_read(tdb, tlock->off, rec) == -1)
-				goto fail;
-			tlock->off = rec->next;
-		}
-
-		/* Iterate through chain */
-		while( tlock->off) {
-			tdb_off current;
-			if (rec_read(tdb, tlock->off, rec) == -1)
-				goto fail;
-
-			/* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
-			if (tlock->off == rec->next) {
-				TDB_LOG((tdb, 0, "tdb_next_lock: loop detected.\n"));
-				goto fail;
-			}
-
-			if (!TDB_DEAD(rec)) {
-				/* Woohoo: we found one! */
-				if (lock_record(tdb, tlock->off) != 0)
-					goto fail;
-				return tlock->off;
-			}
-
-			/* Try to clean dead ones from old traverses */
-			current = tlock->off;
-			tlock->off = rec->next;
-			if (!tdb->read_only && 
-			    do_delete(tdb, current, rec) != 0)
-				goto fail;
-		}
-		tdb_unlock(tdb, tlock->hash, F_WRLCK);
-		want_next = 0;
-	}
-	/* We finished iteration without finding anything */
-	return TDB_ERRCODE(TDB_SUCCESS, 0);
-
- fail:
-	tlock->off = 0;
-	if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0)
-		TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
-	return -1;
-}
-
-/* traverse the entire database - calling fn(tdb, key, data) on each element.
-   return -1 on error or the record count traversed
-   if fn is NULL then it is not called
-   a non-zero return value from fn() indicates that the traversal should stop
-  */
-int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *private_val)
-{
-	TDB_DATA key, dbuf;
-	struct list_struct rec;
-	struct tdb_traverse_lock tl = { NULL, 0, 0 };
-	int ret, count = 0;
-
-	/* This was in the initializaton, above, but the IRIX compiler
-	 * did not like it.  crh
-	 */
-	tl.next = tdb->travlocks.next;
-
-	/* fcntl locks don't stack: beware traverse inside traverse */
-	tdb->travlocks.next = &tl;
-
-	/* tdb_next_lock places locks on the record returned, and its chain */
-	while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
-		count++;
-		/* now read the full record */
-		key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec), 
-					  rec.key_len + rec.data_len);
-		if (!key.dptr) {
-			ret = -1;
-			if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0)
-				goto out;
-			if (unlock_record(tdb, tl.off) != 0)
-				TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
-			goto out;
-		}
-		key.dsize = rec.key_len;
-		dbuf.dptr = key.dptr + rec.key_len;
-		dbuf.dsize = rec.data_len;
-
-		/* Drop chain lock, call out */
-		if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) {
-			ret = -1;
-			SAFE_FREE(key.dptr);
-			goto out;
-		}
-		if (fn && fn(tdb, key, dbuf, private_val)) {
-			/* They want us to terminate traversal */
-			ret = count;
-			if (unlock_record(tdb, tl.off) != 0) {
-				TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));;
-				ret = -1;
-			}
-			SAFE_FREE(key.dptr);
-			goto out;
-		}
-		SAFE_FREE(key.dptr);
-	}
-out:
-	tdb->travlocks.next = tl.next;
-	if (ret < 0)
-		return -1;
-	else
-		return count;
-}
-
-/* find the first entry in the database and return its key */
-TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
-{
-	TDB_DATA key;
-	struct list_struct rec;
-
-	/* release any old lock */
-	if (unlock_record(tdb, tdb->travlocks.off) != 0)
-		return tdb_null;
-	tdb->travlocks.off = tdb->travlocks.hash = 0;
-
-	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
-		return tdb_null;
-	/* now read the key */
-	key.dsize = rec.key_len;
-	key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
-	if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
-		TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
-	return key;
-}
-
-/* find the next entry in the database, returning its key */
-TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
-{
-	u32 oldhash;
-	TDB_DATA key = tdb_null;
-	struct list_struct rec;
-	char *k = NULL;
-
-	/* Is locked key the old key?  If so, traverse will be reliable. */
-	if (tdb->travlocks.off) {
-		if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
-			return tdb_null;
-		if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
-		    || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
-					    rec.key_len))
-		    || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
-			/* No, it wasn't: unlock it and start from scratch */
-			if (unlock_record(tdb, tdb->travlocks.off) != 0) {
-				SAFE_FREE(k);
-				return tdb_null;
-			}
-			if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0) {
-				SAFE_FREE(k);
-				return tdb_null;
-			}
-			tdb->travlocks.off = 0;
-		}
-
-		SAFE_FREE(k);
-	}
-
-	if (!tdb->travlocks.off) {
-		/* No previous element: do normal find, and lock record */
-		tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), F_WRLCK, &rec);
-		if (!tdb->travlocks.off)
-			return tdb_null;
-		tdb->travlocks.hash = BUCKET(rec.full_hash);
-		if (lock_record(tdb, tdb->travlocks.off) != 0) {
-			TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
-			return tdb_null;
-		}
-	}
-	oldhash = tdb->travlocks.hash;
-
-	/* Grab next record: locks chain and returned record,
-	   unlocks old record */
-	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
-		key.dsize = rec.key_len;
-		key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
-					  key.dsize);
-		/* Unlock the chain of this new record */
-		if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
-			TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
-	}
-	/* Unlock the chain of old record */
-	if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
-		TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
-	return key;
-}
-
 /* delete an entry in the database given a key */
-static int tdb_delete_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
+static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash)
 {
-	tdb_off rec_ptr;
+	tdb_off_t rec_ptr;
 	struct list_struct rec;
 	int ret;
 
 	if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec)))
 		return -1;
-	ret = do_delete(tdb, rec_ptr, &rec);
+	ret = tdb_do_delete(tdb, rec_ptr, &rec);
+
+	if (ret == 0) {
+		tdb_increment_seqnum(tdb);
+	}
+
 	if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
 		TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
 	return ret;
 }
 
-int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
+int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
 {
 	u32 hash = tdb->hash_fn(&key);
 	return tdb_delete_hash(tdb, key, hash);
@@ -1527,14 +251,19 @@ int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
 
    return 0 on success, -1 on failure
 */
-int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
+int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 {
 	struct list_struct rec;
 	u32 hash;
-	tdb_off rec_ptr;
+	tdb_off_t rec_ptr;
 	char *p = NULL;
 	int ret = 0;
 
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
 	/* find which hash bucket it is in */
 	hash = tdb->hash_fn(&key);
 	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
@@ -1555,7 +284,7 @@ int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 			/* if the record doesn't exist and we are in TDB_MODIFY mode then
 			 we should fail the store */
 			goto fail;
-	}
+		}
 	}
 	/* reset the error code potentially set by the tdb_update() */
 	tdb->ecode = TDB_SUCCESS;
@@ -1583,7 +312,7 @@ int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 		goto fail;
 
 	/* Read hash top into next ptr */
-	if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
 		goto fail;
 
 	rec.key_len = key.dsize;
@@ -1592,12 +321,15 @@ int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 	rec.magic = TDB_MAGIC;
 
 	/* write out and point the top of the hash chain at it */
-	if (rec_write(tdb, rec_ptr, &rec) == -1
-	    || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
-	    || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
+	if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
+	    || tdb->methods->tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
+	    || tdb_ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
 		/* Need to tdb_unallocate() here */
 		goto fail;
 	}
+
+	tdb_increment_seqnum(tdb);
+
  out:
 	SAFE_FREE(p); 
 	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
@@ -1607,521 +339,92 @@ fail:
 	goto out;
 }
 
-/* Attempt to append data to an entry in place - this only works if the new data size
-   is <= the old data size and the key exists.
-   on failure return -1. Record must be locked before calling.
-*/
-static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA new_dbuf)
-{
-	struct list_struct rec;
-	tdb_off rec_ptr;
-
-	/* find entry */
-	if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
-		return -1;
-
-	/* Append of 0 is always ok. */
-	if (new_dbuf.dsize == 0)
-		return 0;
-
-	/* must be long enough for key, old data + new data and tailer */
-	if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + sizeof(tdb_off)) {
-		/* No room. */
-		tdb->ecode = TDB_SUCCESS; /* Not really an error */
-		return -1;
-	}
-
-	if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len,
-		      new_dbuf.dptr, new_dbuf.dsize) == -1)
-		return -1;
-
-	/* update size */
-	rec.data_len += new_dbuf.dsize;
-	return rec_write(tdb, rec_ptr, &rec);
-}
 
 /* Append to an entry. Create if not exist. */
-
-int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
 {
-	struct list_struct rec;
 	u32 hash;
-	tdb_off rec_ptr;
-	char *p = NULL;
-	int ret = 0;
-	size_t new_data_size = 0;
+	TDB_DATA dbuf;
+	int ret = -1;
 
 	/* find which hash bucket it is in */
 	hash = tdb->hash_fn(&key);
 	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
 		return -1;
 
-	/* first try in-place. */
-	if (tdb_append_inplace(tdb, key, hash, new_dbuf) == 0)
-		goto out;
+	dbuf = tdb_fetch(tdb, key);
 
-	/* reset the error code potentially set by the tdb_append_inplace() */
-	tdb->ecode = TDB_SUCCESS;
-
-	/* find entry */
-	if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
-		if (tdb->ecode != TDB_ERR_NOEXIST)
-			goto fail;
-
-		/* Not found - create. */
-
-		ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT);
-		goto out;
+	if (dbuf.dptr == NULL) {
+		dbuf.dptr = malloc(new_dbuf.dsize);
+	} else {
+		dbuf.dptr = realloc(dbuf.dptr, dbuf.dsize + new_dbuf.dsize);
 	}
 
-	new_data_size = rec.data_len + new_dbuf.dsize;
-
-	/* Copy key+old_value+value *before* allocating free space in case malloc
-	   fails and we are left with a dead spot in the tdb. */
-
-	if (!(p = (char *)malloc(key.dsize + new_data_size))) {
+	if (dbuf.dptr == NULL) {
 		tdb->ecode = TDB_ERR_OOM;
-		goto fail;
-	}
-
-	/* Copy the key in place. */
-	memcpy(p, key.dptr, key.dsize);
-
-	/* Now read the old data into place. */
-	if (rec.data_len &&
-		tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + key.dsize, rec.data_len, 0) == -1)
-			goto fail;
-
-	/* Finally append the new data. */
-	if (new_dbuf.dsize)
-		memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize);
-
-	/* delete any existing record - if it doesn't exist we don't
-           care.  Doing this first reduces fragmentation, and avoids
-           coalescing with `allocated' block before it's updated. */
-
-	tdb_delete_hash(tdb, key, hash);
-
-	if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec)))
-		goto fail;
-
-	/* Read hash top into next ptr */
-	if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
-		goto fail;
-
-	rec.key_len = key.dsize;
-	rec.data_len = new_data_size;
-	rec.full_hash = hash;
-	rec.magic = TDB_MAGIC;
-
-	/* write out and point the top of the hash chain at it */
-	if (rec_write(tdb, rec_ptr, &rec) == -1
-	    || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+new_data_size)==-1
-	    || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
-		/* Need to tdb_unallocate() here */
-		goto fail;
+		goto failed;
 	}
 
- out:
-	SAFE_FREE(p); 
-	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
-	return ret;
-
-fail:
-	ret = -1;
-	goto out;
-}
+	memcpy(dbuf.dptr + dbuf.dsize, new_dbuf.dptr, new_dbuf.dsize);
+	dbuf.dsize += new_dbuf.dsize;
 
-static int tdb_already_open(dev_t device,
-			    ino_t ino)
-{
-	TDB_CONTEXT *i;
+	ret = tdb_store(tdb, key, dbuf, 0);
 	
-	for (i = tdbs; i; i = i->next) {
-		if (i->device == device && i->inode == ino) {
-			return 1;
-		}
-	}
-
-	return 0;
-}
-
-/* This is based on the hash algorithm from gdbm */
-static u32 default_tdb_hash(TDB_DATA *key)
-{
-	u32 value;	/* Used to compute the hash value.  */
-	u32   i;	/* Used to cycle through random values. */
-
-	/* Set the initial value from the key size. */
-	for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
-		value = (value + (key->dptr[i] << (i*5 % 24)));
-
-	return (1103515243 * value + 12345);  
-}
-
-/* open the database, creating it if necessary 
-
-   The open_flags and mode are passed straight to the open call on the
-   database file. A flags value of O_WRONLY is invalid. The hash size
-   is advisory, use zero for a default value.
-
-   Return is NULL on error, in which case errno is also set.  Don't 
-   try to call tdb_error or tdb_errname, just do strerror(errno).
-
-   @param name may be NULL for internal databases. */
-TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
-		      int open_flags, mode_t mode)
-{
-	return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
-}
-
-
-TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
-			 int open_flags, mode_t mode,
-			 tdb_log_func log_fn,
-			 tdb_hash_func hash_fn)
-{
-	TDB_CONTEXT *tdb;
-	struct stat st;
-	int rev = 0, locked = 0;
-	unsigned char *vp;
-	u32 vertest;
-
-	if (!(tdb = calloc(1, sizeof *tdb))) {
-		/* Can't log this */
-		errno = ENOMEM;
-		goto fail;
-	}
-	tdb->fd = -1;
-	tdb->name = NULL;
-	tdb->map_ptr = NULL;
-	tdb->flags = tdb_flags;
-	tdb->open_flags = open_flags;
-	tdb->log_fn = log_fn;
-	tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
-
-	if ((open_flags & O_ACCMODE) == O_WRONLY) {
-		TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
-			 name));
-		errno = EINVAL;
-		goto fail;
-	}
-	
-	if (hash_size == 0)
-		hash_size = DEFAULT_HASH_SIZE;
-	if ((open_flags & O_ACCMODE) == O_RDONLY) {
-		tdb->read_only = 1;
-		/* read only databases don't do locking or clear if first */
-		tdb->flags |= TDB_NOLOCK;
-		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
-	}
-
-	/* internal databases don't mmap or lock, and start off cleared */
-	if (tdb->flags & TDB_INTERNAL) {
-		tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
-		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
-		if (tdb_new_database(tdb, hash_size) != 0) {
-			TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!"));
-			goto fail;
-		}
-		goto internal;
-	}
-
-	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
-		TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
-			 name, strerror(errno)));
-		goto fail;	/* errno set by open(2) */
-	}
-
-	/* ensure there is only one process initialising at once */
-	if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
-		TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
-			 name, strerror(errno)));
-		goto fail;	/* errno set by tdb_brlock */
-	}
-
-	/* we need to zero database if we are the only one with it open */
-	if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
-		(locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))) {
-		open_flags |= O_CREAT;
-		if (ftruncate(tdb->fd, 0) == -1) {
-			TDB_LOG((tdb, 0, "tdb_open_ex: "
-				 "failed to truncate %s: %s\n",
-				 name, strerror(errno)));
-			goto fail; /* errno set by ftruncate */
-		}
-	}
-
-	if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
-	    || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
-	    || (tdb->header.version != TDB_VERSION
-		&& !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
-		/* its not a valid database - possibly initialise it */
-		if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
-			errno = EIO; /* ie bad format or something */
-			goto fail;
-		}
-		rev = (tdb->flags & TDB_CONVERT);
-	}
-	vp = (unsigned char *)&tdb->header.version;
-	vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
-		  (((u32)vp[2]) << 8) | (u32)vp[3];
-	tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
-	if (!rev)
-		tdb->flags &= ~TDB_CONVERT;
-	else {
-		tdb->flags |= TDB_CONVERT;
-		convert(&tdb->header, sizeof(tdb->header));
-	}
-	if (fstat(tdb->fd, &st) == -1)
-		goto fail;
-
-	/* Is it already in the open list?  If so, fail. */
-	if (tdb_already_open(st.st_dev, st.st_ino)) {
-		TDB_LOG((tdb, 2, "tdb_open_ex: "
-			 "%s (%d,%d) is already open in this process\n",
-			 name, (int)st.st_dev, (int)st.st_ino));
-		errno = EBUSY;
-		goto fail;
-	}
-
-	if (!(tdb->name = (char *)strdup(name))) {
-		errno = ENOMEM;
-		goto fail;
-	}
-
-	tdb->map_size = st.st_size;
-	tdb->device = st.st_dev;
-	tdb->inode = st.st_ino;
-	tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
-	if (!tdb->locked) {
-		TDB_LOG((tdb, 2, "tdb_open_ex: "
-			 "failed to allocate lock structure for %s\n",
-			 name));
-		errno = ENOMEM;
-		goto fail;
-	}
-	tdb_mmap(tdb);
-	if (locked) {
-		if (!tdb->read_only)
-			if (tdb_clear_spinlocks(tdb) != 0) {
-				TDB_LOG((tdb, 0, "tdb_open_ex: "
-				"failed to clear spinlock\n"));
-				goto fail;
-			}
-		if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
-			TDB_LOG((tdb, 0, "tdb_open_ex: "
-				 "failed to take ACTIVE_LOCK on %s: %s\n",
-				 name, strerror(errno)));
-			goto fail;
-		}
-
-	}
-
-	/* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
-	   we didn't get the initial exclusive lock as we need to let all other
-	   users know we're using it. */
-
-	if (tdb_flags & TDB_CLEAR_IF_FIRST) {
-		/* leave this lock in place to indicate it's in use */
-		if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
-			goto fail;
-	}
-
-
- internal:
-	/* Internal (memory-only) databases skip all the code above to
-	 * do with disk files, and resume here by releasing their
-	 * global lock and hooking into the active list. */
-	if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
-		goto fail;
-	tdb->next = tdbs;
-	tdbs = tdb;
-	return tdb;
-
- fail:
-	{ int save_errno = errno;
-
-	if (!tdb)
-		return NULL;
-	
-	if (tdb->map_ptr) {
-		if (tdb->flags & TDB_INTERNAL)
-			SAFE_FREE(tdb->map_ptr);
-		else
-			tdb_munmap(tdb);
-	}
-	SAFE_FREE(tdb->name);
-	if (tdb->fd != -1)
-		if (close(tdb->fd) != 0)
-			TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
-	SAFE_FREE(tdb->locked);
-	SAFE_FREE(tdb);
-	errno = save_errno;
-	return NULL;
-	}
-}
-
-/**
- * Close a database.
- *
- * @returns -1 for error; 0 for success.
- **/
-int tdb_close(TDB_CONTEXT *tdb)
-{
-	TDB_CONTEXT **i;
-	int ret = 0;
-
-	if (tdb->map_ptr) {
-		if (tdb->flags & TDB_INTERNAL)
-			SAFE_FREE(tdb->map_ptr);
-		else
-			tdb_munmap(tdb);
-	}
-	SAFE_FREE(tdb->name);
-	if (tdb->fd != -1)
-		ret = close(tdb->fd);
-	SAFE_FREE(tdb->locked);
-
-	/* Remove from contexts list */
-	for (i = &tdbs; *i; i = &(*i)->next) {
-		if (*i == tdb) {
-			*i = tdb->next;
-			break;
-		}
-	}
-
-	memset(tdb, 0, sizeof(*tdb));
-	SAFE_FREE(tdb);
-
+failed:
+	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+	SAFE_FREE(dbuf.dptr);
 	return ret;
 }
 
-/* lock/unlock entire database */
-int tdb_lockall(TDB_CONTEXT *tdb)
-{
-	u32 i;
-
-	/* There are no locks on read-only dbs */
-	if (tdb->read_only)
-		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
-	for (i = 0; i < tdb->header.hash_size; i++) 
-		if (tdb_lock(tdb, i, F_WRLCK))
-			break;
-
-	/* If error, release locks we have... */
-	if (i < tdb->header.hash_size) {
-		u32 j;
-
-		for ( j = 0; j < i; j++)
-			tdb_unlock(tdb, j, F_WRLCK);
-		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
-	}
-
-	return 0;
-}
-void tdb_unlockall(TDB_CONTEXT *tdb)
-{
-	u32 i;
-	for (i=0; i < tdb->header.hash_size; i++)
-		tdb_unlock(tdb, i, F_WRLCK);
-}
-
-/* lock/unlock one hash chain. This is meant to be used to reduce
-   contention - it cannot guarantee how many records will be locked */
-int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
-{
-	return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
-}
 
-int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
+/*
+  return the name of the current tdb file
+  useful for external logging functions
+*/
+const char *tdb_name(struct tdb_context *tdb)
 {
-	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+	return tdb->name;
 }
 
-int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+/*
+  return the underlying file descriptor being used by tdb, or -1
+  useful for external routines that want to check the device/inode
+  of the fd
+*/
+int tdb_fd(struct tdb_context *tdb)
 {
-	return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+	return tdb->fd;
 }
 
-int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+/*
+  return the current logging function
+  useful for external tdb routines that wish to log tdb errors
+*/
+tdb_log_func tdb_log_fn(struct tdb_context *tdb)
 {
-	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+	return tdb->log_fn;
 }
 
 
-/* register a loging function */
-void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))
-{
-	tdb->log_fn = fn;
-}
-
-/* reopen a tdb - this can be used after a fork to ensure that we have an independent
-   seek pointer from our parent and to re-establish locks */
-int tdb_reopen(TDB_CONTEXT *tdb)
+/*
+  get the tdb sequence number. Only makes sense if the writers opened
+  with TDB_SEQNUM set. Note that this sequence number will wrap quite
+  quickly, so it should only be used for a 'has something changed'
+  test, not for code that relies on the count of the number of changes
+  made. If you want a counter then use a tdb record.
+
+  The aim of this sequence number is to allow for a very lightweight
+  test of a possible tdb change.
+*/
+int tdb_get_seqnum(struct tdb_context *tdb)
 {
-	struct stat st;
+	tdb_off_t seqnum=0;
 
-	if (tdb->flags & TDB_INTERNAL)
-		return 0; /* Nothing to do. */
-	if (tdb_munmap(tdb) != 0) {
-		TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
-		goto fail;
-	}
-	if (close(tdb->fd) != 0)
-		TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
-	tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
-	if (tdb->fd == -1) {
-		TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
-		goto fail;
-	}
-	if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
-			(tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)) {
-		TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
-		goto fail;
-	}
-	if (fstat(tdb->fd, &st) != 0) {
-		TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
-		goto fail;
-	}
-	if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
-		TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
-		goto fail;
-	}
-	tdb_mmap(tdb);
-
-	return 0;
-
-fail:
-	tdb_close(tdb);
-	return -1;
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	return seqnum;
 }
 
-/* reopen all tdb's */
-int tdb_reopen_all(int parent_longlived)
+int tdb_hash_size(struct tdb_context *tdb)
 {
-	TDB_CONTEXT *tdb;
-
-	for (tdb=tdbs; tdb; tdb = tdb->next) {
-		/*
-		 * If the parent is longlived (ie. a
-		 * parent daemon architecture), we know
-		 * it will keep it's active lock on a
-		 * tdb opened with CLEAR_IF_FIRST. Thus
-		 * for child processes we don't have to
-		 * add an active lock. This is essential
-		 * to improve performance on systems that
-		 * keep POSIX locks as a non-scalable data
-		 * structure in the kernel.
-		 */
-		if (parent_longlived) {
-			/* Ensure no clear-if-first. */
-			tdb->flags &= ~TDB_CLEAR_IF_FIRST;
-		}
-
-		if (tdb_reopen(tdb) != 0)
-			return -1;
-	}
-
-	return 0;
+	return tdb->header.hash_size;
 }
diff --git a/source3/tdb/tdb.h b/source3/tdb/tdb.h
index b5b87ee5a5..82f8828c89 100644
--- a/source3/tdb/tdb.h
+++ b/source3/tdb/tdb.h
@@ -3,9 +3,9 @@
 
 /* 
    Unix SMB/CIFS implementation.
-   
+
    trivial database library
-   
+
    Copyright (C) Andrew Tridgell 1999-2004
    
      ** NOTE! The following LGPL license applies to the tdb
@@ -21,7 +21,7 @@
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Lesser General Public License for more details.
-   
+
    You should have received a copy of the GNU Lesser General Public
    License along with this library; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
@@ -31,17 +31,6 @@
 extern "C" {
 #endif
 
-#ifndef PRINTF_ATTRIBUTE
-/** Use gcc attribute to check printf fns.  a1 is the 1-based index of
- * the parameter containing the format, and a2 the index of the first
- * argument. Note that some gcc 2.x versions don't handle this
- * properly **/
-#if (__GNUC__ >= 3) && (__GNUC_MINOR__ >= 1 )
-#define PRINTF_ATTRIBUTE(a1, a2) __attribute__ ((format (__printf__, a1, a2)))
-#else
-#define PRINTF_ATTRIBUTE(a1, a2)
-#endif
-#endif
 
 /* flags to tdb_store() */
 #define TDB_REPLACE 1
@@ -56,102 +45,83 @@ extern "C" {
 #define TDB_NOMMAP   8 /* don't use mmap */
 #define TDB_CONVERT 16 /* convert endian (internal use) */
 #define TDB_BIGENDIAN 32 /* header is big-endian (internal use) */
+#define TDB_NOSYNC   64 /* don't use synchronous transactions */
+#define TDB_SEQNUM   128 /* maintain a sequence number */
 
 #define TDB_ERRCODE(code, ret) ((tdb->ecode = (code)), ret)
 
 /* error codes */
 enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK, 
 		TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOLOCK, TDB_ERR_LOCK_TIMEOUT,
-		TDB_ERR_NOEXIST};
-
-#ifndef u32
-#define u32 unsigned
-#endif
+		TDB_ERR_NOEXIST, TDB_ERR_EINVAL, TDB_ERR_RDONLY};
 
-typedef struct {
+typedef struct TDB_DATA {
 	char *dptr;
 	size_t dsize;
 } TDB_DATA;
 
-typedef u32 tdb_len;
-typedef u32 tdb_off;
-
-/* this is stored at the front of every database */
-struct tdb_header {
-	char magic_food[32]; /* for /etc/magic */
-	u32 version; /* version of the code */
-	u32 hash_size; /* number of hash entries */
-	tdb_off rwlocks;
-	tdb_off reserved[31];
-};
-
-struct tdb_lock_type {
-	u32 count;
-	u32 ltype;
-};
-
-struct tdb_traverse_lock {
-	struct tdb_traverse_lock *next;
-	u32 off;
-	u32 hash;
-};
+#ifndef PRINTF_ATTRIBUTE
+#if (__GNUC__ >= 3)
+/** Use gcc attribute to check printf fns.  a1 is the 1-based index of
+ * the parameter containing the format, and a2 the index of the first
+ * argument. Note that some gcc 2.x versions don't handle this
+ * properly **/
+#define PRINTF_ATTRIBUTE(a1, a2) __attribute__ ((format (__printf__, a1, a2)))
+#else
+#define PRINTF_ATTRIBUTE(a1, a2)
+#endif
+#endif
 
 /* this is the context structure that is returned from a db open */
-typedef struct tdb_context {
-	char *name; /* the name of the database */
-	void *map_ptr; /* where it is currently mapped */
-	int fd; /* open file descriptor for the database */
-	tdb_len map_size; /* how much space has been mapped */
-	int read_only; /* opened read-only */
-	struct tdb_lock_type *locked; /* array of chain locks */
-	enum TDB_ERROR ecode; /* error code for last tdb error */
-	struct tdb_header header; /* a cached copy of the header */
-	u32 flags; /* the flags passed to tdb_open */
-	struct tdb_traverse_lock travlocks; /* current traversal locks */
-	struct tdb_context *next; /* all tdbs to avoid multiple opens */
-	dev_t device;	/* uniquely identifies this tdb */
-	ino_t inode;	/* uniquely identifies this tdb */
-	void (*log_fn)(struct tdb_context *tdb, int level, const char *, ...) PRINTF_ATTRIBUTE(3,4); /* logging function */
-	u32 (*hash_fn)(TDB_DATA *key);
-	int open_flags; /* flags used in the open - needed by reopen */
-} TDB_CONTEXT;
-
-typedef int (*tdb_traverse_func)(TDB_CONTEXT *, TDB_DATA, TDB_DATA, void *);
-typedef void (*tdb_log_func)(TDB_CONTEXT *, int , const char *, ...);
-typedef u32 (*tdb_hash_func)(TDB_DATA *key);
-
-TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
+typedef struct tdb_context TDB_CONTEXT;
+
+typedef int (*tdb_traverse_func)(struct tdb_context *, TDB_DATA, TDB_DATA, void *);
+typedef void (*tdb_log_func)(struct tdb_context *, int , const char *, ...);
+typedef unsigned int (*tdb_hash_func)(TDB_DATA *key);
+
+struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
 		      int open_flags, mode_t mode);
-TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 			 int open_flags, mode_t mode,
 			 tdb_log_func log_fn,
 			 tdb_hash_func hash_fn);
 
-int tdb_reopen(TDB_CONTEXT *tdb);
-int tdb_reopen_all(int);
-void tdb_logging_function(TDB_CONTEXT *tdb, tdb_log_func);
-enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb);
-const char *tdb_errorstr(TDB_CONTEXT *tdb);
-TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
-int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf);
-int tdb_close(TDB_CONTEXT *tdb);
-TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb);
-TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *);
-int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_lockall(TDB_CONTEXT *tdb);
-void tdb_unlockall(TDB_CONTEXT *tdb);
+int tdb_reopen(struct tdb_context *tdb);
+int tdb_reopen_all(int parent_longlived);
+void tdb_logging_function(struct tdb_context *tdb, tdb_log_func);
+enum TDB_ERROR tdb_error(struct tdb_context *tdb);
+const char *tdb_errorstr(struct tdb_context *tdb);
+TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
+int tdb_delete(struct tdb_context *tdb, TDB_DATA key);
+int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
+int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf);
+int tdb_close(struct tdb_context *tdb);
+TDB_DATA tdb_firstkey(struct tdb_context *tdb);
+TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA key);
+int tdb_traverse(struct tdb_context *tdb, tdb_traverse_func fn, void *);
+int tdb_traverse_read(struct tdb_context *tdb, tdb_traverse_func fn, void *);
+int tdb_exists(struct tdb_context *tdb, TDB_DATA key);
+int tdb_lockall(struct tdb_context *tdb);
+void tdb_unlockall(struct tdb_context *tdb);
+const char *tdb_name(struct tdb_context *tdb);
+int tdb_fd(struct tdb_context *tdb);
+tdb_log_func tdb_log_fn(struct tdb_context *tdb);
+int tdb_transaction_start(struct tdb_context *tdb);
+int tdb_transaction_commit(struct tdb_context *tdb);
+int tdb_transaction_cancel(struct tdb_context *tdb);
+int tdb_transaction_recover(struct tdb_context *tdb);
+int tdb_get_seqnum(struct tdb_context *tdb);
+int tdb_hash_size(struct tdb_context *tdb);
 
 /* Low level locking functions: use with care */
-void tdb_set_lock_alarm(sig_atomic_t *palarm);
-int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key);
 
 /* Debug functions. Not used in production. */
-void tdb_dump_all(TDB_CONTEXT *tdb);
-int tdb_printfreelist(TDB_CONTEXT *tdb);
+void tdb_dump_all(struct tdb_context *tdb);
+int tdb_printfreelist(struct tdb_context *tdb);
 
 extern TDB_DATA tdb_null;
 
diff --git a/source3/tdb/tdb_private.h b/source3/tdb/tdb_private.h
new file mode 100644
index 0000000000..d10c1c46bf
--- /dev/null
+++ b/source3/tdb/tdb_private.h
@@ -0,0 +1,241 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library - private includes
+
+   Copyright (C) Andrew Tridgell              2005
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#ifndef _SAMBA_BUILD_
+#include "config.h"
+#include <stdlib.h>
+#include <stdio.h>
+#ifdef HAVE_STDINT_H
+#include <stdint.h>
+#endif
+#include <fcntl.h>
+#include <unistd.h>
+#include <string.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#ifdef HAVE_SYS_SELECT_H
+#include <sys/select.h>
+#endif
+#ifdef HAVE_SYS_TIME_H
+#include <sys/time.h>
+#endif
+#include "tdb.h"
+
+#ifndef HAVE_PREAD_DECL
+ssize_t pread(int fd, void *buf, size_t count, off_t offset);
+#endif
+#ifndef HAVE_PWRITE_DECL
+ssize_t pwrite(int fd, const void *buf, size_t count, off_t offset);
+#endif
+
+#else
+#include "includes.h"
+#undef malloc
+#undef realloc
+#undef calloc
+#undef strdup
+#endif
+
+#ifndef u32
+#define u32 unsigned
+#endif
+
+#ifndef HAVE_GETPAGESIZE
+#define getpagesize() 0x2000
+#endif
+
+typedef u32 tdb_len_t;
+typedef u32 tdb_off_t;
+
+#ifndef offsetof
+#define offsetof(t,f) ((unsigned int)&((t *)0)->f)
+#endif
+
+#define TDB_MAGIC_FOOD "TDB file\n"
+#define TDB_VERSION (0x26011967 + 6)
+#define TDB_MAGIC (0x26011999U)
+#define TDB_FREE_MAGIC (~TDB_MAGIC)
+#define TDB_DEAD_MAGIC (0xFEE1DEAD)
+#define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
+#define TDB_ALIGNMENT 4
+#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
+#define DEFAULT_HASH_SIZE 131
+#define FREELIST_TOP (sizeof(struct tdb_header))
+#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
+#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
+#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
+#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
+#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
+#define TDB_HASHTABLE_SIZE(tdb) ((tdb->header.hash_size+1)*sizeof(tdb_off_t))
+#define TDB_DATA_START(hash_size) TDB_HASH_TOP(hash_size-1)
+#define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
+#define TDB_SEQNUM_OFS    offsetof(struct tdb_header, sequence_number)
+#define TDB_PAD_BYTE 0x42
+#define TDB_PAD_U32  0x42424242
+
+/* NB assumes there is a local variable called "tdb" that is the
+ * current context, also takes doubly-parenthesized print-style
+ * argument. */
+#define TDB_LOG(x) tdb->log_fn x
+
+/* lock offsets */
+#define GLOBAL_LOCK      0
+#define ACTIVE_LOCK      4
+#define TRANSACTION_LOCK 8
+
+#ifndef MAP_FILE
+#define MAP_FILE 0
+#endif
+
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+/* free memory if the pointer is valid and zero the pointer */
+#ifndef SAFE_FREE
+#define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
+#endif
+
+#define BUCKET(hash) ((hash) % tdb->header.hash_size)
+
+#define DOCONV() (tdb->flags & TDB_CONVERT)
+#define CONVERT(x) (DOCONV() ? tdb_convert(&x, sizeof(x)) : &x)
+
+
+/* the body of the database is made of one list_struct for the free space
+   plus a separate data list for each hash value */
+struct list_struct {
+	tdb_off_t next; /* offset of the next record in the list */
+	tdb_len_t rec_len; /* total byte length of record */
+	tdb_len_t key_len; /* byte length of key */
+	tdb_len_t data_len; /* byte length of data */
+	u32 full_hash; /* the full 32 bit hash of the key */
+	u32 magic;   /* try to catch errors */
+	/* the following union is implied:
+		union {
+			char record[rec_len];
+			struct {
+				char key[key_len];
+				char data[data_len];
+			}
+			u32 totalsize; (tailer)
+		}
+	*/
+};
+
+
+/* this is stored at the front of every database */
+struct tdb_header {
+	char magic_food[32]; /* for /etc/magic */
+	u32 version; /* version of the code */
+	u32 hash_size; /* number of hash entries */
+	tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
+	tdb_off_t recovery_start; /* offset of transaction recovery region */
+	tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
+	tdb_off_t reserved[29];
+};
+
+struct tdb_lock_type {
+	u32 count;
+	u32 ltype;
+};
+
+struct tdb_traverse_lock {
+	struct tdb_traverse_lock *next;
+	u32 off;
+	u32 hash;
+	int lock_rw;
+};
+
+
+struct tdb_methods {
+	int (*tdb_read)(struct tdb_context *, tdb_off_t , void *, tdb_len_t , int );
+	int (*tdb_write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
+	void (*next_hash_chain)(struct tdb_context *, u32 *);
+	int (*tdb_oob)(struct tdb_context *, tdb_off_t , int );
+	int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
+	int (*tdb_brlock)(struct tdb_context *, tdb_off_t , int, int, int);
+};
+
+struct tdb_context {
+	char *name; /* the name of the database */
+	void *map_ptr; /* where it is currently mapped */
+	int fd; /* open file descriptor for the database */
+	tdb_len_t map_size; /* how much space has been mapped */
+	int read_only; /* opened read-only */
+	int traverse_read; /* read-only traversal */
+	struct tdb_lock_type *locked; /* array of chain locks */
+	enum TDB_ERROR ecode; /* error code for last tdb error */
+	struct tdb_header header; /* a cached copy of the header */
+	u32 flags; /* the flags passed to tdb_open */
+	struct tdb_traverse_lock travlocks; /* current traversal locks */
+	struct tdb_context *next; /* all tdbs to avoid multiple opens */
+	dev_t device;	/* uniquely identifies this tdb */
+	ino_t inode;	/* uniquely identifies this tdb */
+	void (*log_fn)(struct tdb_context *tdb, int level, const char *, ...) PRINTF_ATTRIBUTE(3,4); /* logging function */
+	unsigned int (*hash_fn)(TDB_DATA *key);
+	int open_flags; /* flags used in the open - needed by reopen */
+	unsigned int num_locks; /* number of chain locks held */
+	const struct tdb_methods *methods;
+	struct tdb_transaction *transaction;
+	int page_size;
+};
+
+
+/*
+  internal prototypes
+*/
+int tdb_munmap(struct tdb_context *tdb);
+void tdb_mmap(struct tdb_context *tdb);
+int tdb_lock(struct tdb_context *tdb, int list, int ltype);
+int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, int rw_type, int lck_type, int probe);
+int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len);
+int tdb_brlock_len(struct tdb_context *tdb, tdb_off_t offset, 
+		   int rw_type, int lck_type, int probe, size_t len);
+int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+void *tdb_convert(void *buf, u32 size);
+int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec);
+tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct list_struct *rec);
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec);
+int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec);
+int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct list_struct *rec);
+char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash, int locktype,
+			   struct list_struct *rec);
+void tdb_io_init(struct tdb_context *tdb);
+int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
+
+
diff --git a/source3/tdb/tdbback.c b/source3/tdb/tdbback.c
index 7967a6bbf5..c337924d6c 100644
--- a/source3/tdb/tdbback.c
+++ b/source3/tdb/tdbback.c
@@ -77,7 +77,7 @@ static int copy_fn(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
 	TDB_CONTEXT *tdb_new = (TDB_CONTEXT *)state;
 
 	if (tdb_store(tdb_new, key, dbuf, TDB_INSERT) != 0) {
-		fprintf(stderr,"Failed to insert into %s\n", tdb_new->name);
+		fprintf(stderr,"Failed to insert into %s\n", tdb_name(tdb));
 		failed = 1;
 		return 1;
 	}
@@ -122,7 +122,7 @@ int backup_tdb(const char *old_name, const char *new_name)
 
 	/* create the new tdb */
 	unlink(tmp_name);
-	tdb_new = tdb_open(tmp_name, tdb->header.hash_size, 
+	tdb_new = tdb_open(tmp_name, tdb_hash_size(tdb),
 			   TDB_DEFAULT, O_RDWR|O_CREAT|O_EXCL, 
 			   st.st_mode & 0777);
 	if (!tdb_new) {
@@ -179,7 +179,7 @@ int backup_tdb(const char *old_name, const char *new_name)
 	}
 
 	/* make sure the new tdb has reached stable storage */
-	fsync(tdb_new->fd);
+	fsync(tdb_fd(tdb_new));
 
 	/* close the new tdb and rename it to .bak */
 	tdb_close(tdb_new);
diff --git a/source3/tdb/tdbutil.c b/source3/tdb/tdbutil.c
index fc45fa9360..e13f7e3e67 100644
--- a/source3/tdb/tdbutil.c
+++ b/source3/tdb/tdbutil.c
@@ -19,9 +19,21 @@
    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
-#include "includes.h"
+#include "tdb_private.h"
 #include <fnmatch.h>
 
+/***************************************************************
+ Allow a caller to set a "alarm" flag that tdb can check to abort
+ a blocking lock on SIGALRM.
+***************************************************************/
+
+static sig_atomic_t *palarm_fired;
+
+static void tdb_set_lock_alarm(sig_atomic_t *palarm)
+{
+	palarm_fired = palarm;
+}
+
 /* these are little tdb utility functions that are meant to make
    dealing with a tdb database a little less cumbersome in Samba */
 
@@ -641,129 +653,6 @@ int tdb_unpack(char *buf, int bufsize, const char *fmt, ...)
 }
 
 
-/**
- * Pack SID passed by pointer
- *
- * @param pack_buf pointer to buffer which is to be filled with packed data
- * @param bufsize size of packing buffer
- * @param sid pointer to sid to be packed
- *
- * @return length of the packed representation of the whole structure
- **/
-size_t tdb_sid_pack(char* pack_buf, int bufsize, DOM_SID* sid)
-{
-	int idx;
-	size_t len = 0;
-	
-	if (!sid || !pack_buf) return -1;
-	
-	len += tdb_pack(pack_buf + len, bufsize - len, "bb", sid->sid_rev_num,
-	                sid->num_auths);
-	
-	for (idx = 0; idx < 6; idx++) {
-		len += tdb_pack(pack_buf + len, bufsize - len, "b", sid->id_auth[idx]);
-	}
-	
-	for (idx = 0; idx < MAXSUBAUTHS; idx++) {
-		len += tdb_pack(pack_buf + len, bufsize - len, "d", sid->sub_auths[idx]);
-	}
-	
-	return len;
-}
-
-
-/**
- * Unpack SID into a pointer
- *
- * @param pack_buf pointer to buffer with packed representation
- * @param bufsize size of the buffer
- * @param sid pointer to sid structure to be filled with unpacked data
- *
- * @return size of structure unpacked from buffer
- **/
-size_t tdb_sid_unpack(char* pack_buf, int bufsize, DOM_SID* sid)
-{
-	int idx, len = 0;
-	
-	if (!sid || !pack_buf) return -1;
-
-	len += tdb_unpack(pack_buf + len, bufsize - len, "bb",
-	                  &sid->sid_rev_num, &sid->num_auths);
-			  
-	for (idx = 0; idx < 6; idx++) {
-		len += tdb_unpack(pack_buf + len, bufsize - len, "b", &sid->id_auth[idx]);
-	}
-	
-	for (idx = 0; idx < MAXSUBAUTHS; idx++) {
-		len += tdb_unpack(pack_buf + len, bufsize - len, "d", &sid->sub_auths[idx]);
-	}
-	
-	return len;
-}
-
-
-/**
- * Pack TRUSTED_DOM_PASS passed by pointer
- *
- * @param pack_buf pointer to buffer which is to be filled with packed data
- * @param bufsize size of the buffer
- * @param pass pointer to trusted domain password to be packed
- *
- * @return length of the packed representation of the whole structure
- **/
-size_t tdb_trusted_dom_pass_pack(char* pack_buf, int bufsize, TRUSTED_DOM_PASS* pass)
-{
-	int idx, len = 0;
-	
-	if (!pack_buf || !pass) return -1;
-	
-	/* packing unicode domain name and password */
-	len += tdb_pack(pack_buf + len, bufsize - len, "d", pass->uni_name_len);
-	
-	for (idx = 0; idx < 32; idx++)
-		len +=  tdb_pack(pack_buf + len, bufsize - len, "w", pass->uni_name[idx]);
-	
-	len += tdb_pack(pack_buf + len, bufsize - len, "dPd", pass->pass_len,
-	                     pass->pass, pass->mod_time);
-
-	/* packing SID structure */
-	len += tdb_sid_pack(pack_buf + len, bufsize - len, &pass->domain_sid);
-
-	return len;
-}
-
-
-/**
- * Unpack TRUSTED_DOM_PASS passed by pointer
- *
- * @param pack_buf pointer to buffer with packed representation
- * @param bufsize size of the buffer
- * @param pass pointer to trusted domain password to be filled with unpacked data
- *
- * @return size of structure unpacked from buffer
- **/
-size_t tdb_trusted_dom_pass_unpack(char* pack_buf, int bufsize, TRUSTED_DOM_PASS* pass)
-{
-	int idx, len = 0;
-	
-	if (!pack_buf || !pass) return -1;
-
-	/* unpack unicode domain name and plaintext password */
-	len += tdb_unpack(pack_buf, bufsize - len, "d", &pass->uni_name_len);
-	
-	for (idx = 0; idx < 32; idx++)
-		len +=  tdb_unpack(pack_buf + len, bufsize - len, "w", &pass->uni_name[idx]);
-
-	len += tdb_unpack(pack_buf + len, bufsize - len, "dPd", &pass->pass_len, &pass->pass,
-	                  &pass->mod_time);
-	
-	/* unpack domain sid */
-	len += tdb_sid_unpack(pack_buf + len, bufsize - len, &pass->domain_sid);
-	
-	return len;	
-}
-
-
 /****************************************************************************
  Log tdb messages via DEBUG().
 ****************************************************************************/
@@ -806,15 +695,6 @@ TDB_CONTEXT *tdb_open_log(const char *name, int hash_size, int tdb_flags,
 }
 
 /****************************************************************************
-  return the name of the current tdb file useful for external logging
-  functions
-****************************************************************************/
-const char *tdb_name(struct tdb_context *tdb)
-{
-	return tdb->name;
-}
-
-/****************************************************************************
  Allow tdb_delete to be used as a tdb_traversal_fn.
 ****************************************************************************/
 
@@ -893,3 +773,43 @@ void tdb_search_list_free(TDB_LIST_NODE* node)
 		node = next_node;
 	};
 }
+
+size_t tdb_map_size(struct tdb_context *tdb)
+{
+	return tdb->map_size;
+}
+
+int tdb_get_flags(struct tdb_context *tdb)
+{
+	return tdb->flags;
+}
+
+/****************************************************************************
+ tdb_store, wrapped in a transaction. This way we make sure that a process
+ that dies within writing does not leave a corrupt tdb behind.
+****************************************************************************/
+
+int tdb_trans_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+		    int flag)
+{
+	int res;
+
+	if ((res = tdb_transaction_start(tdb)) != 0) {
+		DEBUG(5, ("tdb_transaction_start failed\n"));
+		return res;
+	}
+
+	if ((res = tdb_store(tdb, key, dbuf, flag)) != 0) {
+		DEBUG(10, ("tdb_store failed\n"));
+		if (tdb_transaction_cancel(tdb) != 0) {
+			smb_panic("Cancelling transaction failed\n");
+		}
+		return res;
+	}
+
+	if ((res = tdb_transaction_commit(tdb)) != 0) {
+		DEBUG(5, ("tdb_transaction_commit failed\n"));
+	}
+
+	return res;
+}
diff --git a/source3/tdb/tdbutil.h b/source3/tdb/tdbutil.h
index 1a181a962f..44351619f4 100644
--- a/source3/tdb/tdbutil.h
+++ b/source3/tdb/tdbutil.h
@@ -1,8 +1,7 @@
 /* 
    Unix SMB/CIFS implementation.
    tdb utility functions
-   Copyright (C) Andrew Tridgell   1999
-   Copyright (C) Rafal Szczesniak  2002
+   Copyright (C) Andrew Tridgell 1999
    
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -22,6 +21,7 @@
 #ifndef __TDBUTIL_H__
 #define __TDBUTIL_H__
 
+#include "tdb.h"
 
 /* single node of a list returned by tdb_search_keys */
 typedef struct keys_node 
@@ -31,8 +31,38 @@ typedef struct keys_node
 } TDB_LIST_NODE;
 
 
-TDB_LIST_NODE *tdb_search_keys(TDB_CONTEXT*, const char*);
+TDB_LIST_NODE *tdb_search_keys(struct tdb_context*, const char*);
 void tdb_search_list_free(TDB_LIST_NODE*);
-
+int32_t tdb_change_int32_atomic(struct tdb_context *tdb, const char *keystr, int32_t *oldval, int32_t change_val);
+int tdb_lock_bystring(struct tdb_context *tdb, const char *keyval);
+int tdb_lock_bystring_with_timeout(TDB_CONTEXT *tdb, const char *keyval,
+				   int timeout);
+void tdb_unlock_bystring(struct tdb_context *tdb, const char *keyval);
+int tdb_read_lock_bystring_with_timeout(TDB_CONTEXT *tdb, const char *keyval,
+					unsigned int timeout);
+void tdb_read_unlock_bystring(TDB_CONTEXT *tdb, const char *keyval);
+int32_t tdb_fetch_int32(struct tdb_context *tdb, const char *keystr);
+BOOL tdb_store_uint32(struct tdb_context *tdb, const char *keystr, uint32_t value);
+int tdb_store_int32(struct tdb_context *tdb, const char *keystr, int32_t v);
+BOOL tdb_fetch_uint32(struct tdb_context *tdb, const char *keystr, uint32_t *value);
+int tdb_traverse_delete_fn(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf,
+                     void *state);
+int tdb_store_bystring(struct tdb_context *tdb, const char *keystr, TDB_DATA data, int flags);
+TDB_DATA tdb_fetch_bystring(struct tdb_context *tdb, const char *keystr);
+int tdb_delete_bystring(struct tdb_context *tdb, const char *keystr);
+struct tdb_context *tdb_open_log(const char *name, int hash_size,
+				 int tdb_flags, int open_flags, mode_t mode);
+int tdb_unpack(char *buf, int bufsize, const char *fmt, ...);
+size_t tdb_pack(char *buf, int bufsize, const char *fmt, ...);
+TDB_DATA make_tdb_data(const char *dptr, size_t dsize);
+TDB_DATA string_tdb_data(const char *string);
+int tdb_trans_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+		    int flag);
+BOOL tdb_change_uint32_atomic(TDB_CONTEXT *tdb, const char *keystr,
+			      uint32 *oldval, uint32 change_val);
+int tdb_chainlock_with_timeout( TDB_CONTEXT *tdb, TDB_DATA key,
+				unsigned int timeout);
+int tdb_get_flags(struct tdb_context *tdb);
+size_t tdb_map_size(struct tdb_context *tdb);
 
 #endif /* __TDBUTIL_H__ */
diff --git a/source3/tdb/transaction.c b/source3/tdb/transaction.c
new file mode 100644
index 0000000000..aa54c4ae3c
--- /dev/null
+++ b/source3/tdb/transaction.c
@@ -0,0 +1,1043 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              2005
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+/*
+  transaction design:
+
+  - only allow a single transaction at a time per database. This makes
+    using the transaction API simpler, as otherwise the caller would
+    have to cope with temporary failures in transactions that conflict
+    with other current transactions
+
+  - keep the transaction recovery information in the same file as the
+    database, using a special 'transaction recovery' record pointed at
+    by the header. This removes the need for extra journal files as
+    used by some other databases
+
+  - dymacially allocated the transaction recover record, re-using it
+    for subsequent transactions. If a larger record is needed then
+    tdb_free() the old record to place it on the normal tdb freelist
+    before allocating the new record
+
+  - during transactions, keep a linked list of writes all that have
+    been performed by intercepting all tdb_write() calls. The hooked
+    transaction versions of tdb_read() and tdb_write() check this
+    linked list and try to use the elements of the list in preference
+    to the real database.
+
+  - don't allow any locks to be held when a transaction starts,
+    otherwise we can end up with deadlock (plus lack of lock nesting
+    in posix locks would mean the lock is lost)
+
+  - if the caller gains a lock during the transaction but doesn't
+    release it then fail the commit
+
+  - allow for nested calls to tdb_transaction_start(), re-using the
+    existing transaction record. If the inner transaction is cancelled
+    then a subsequent commit will fail
+ 
+  - keep a mirrored copy of the tdb hash chain heads to allow for the
+    fast hash heads scan on traverse, updating the mirrored copy in
+    the transaction version of tdb_write
+
+  - allow callers to mix transaction and non-transaction use of tdb,
+    although once a transaction is started then an exclusive lock is
+    gained until the transaction is committed or cancelled
+
+  - the commit stategy involves first saving away all modified data
+    into a linearised buffer in the transaction recovery area, then
+    marking the transaction recovery area with a magic value to
+    indicate a valid recovery record. In total 4 fsync/msync calls are
+    needed per commit to prevent race conditions. It might be possible
+    to reduce this to 3 or even 2 with some more work.
+
+  - check for a valid recovery record on open of the tdb, while the
+    global lock is held. Automatically recover from the transaction
+    recovery area if needed, then continue with the open as
+    usual. This allows for smooth crash recovery with no administrator
+    intervention.
+
+  - if TDB_NOSYNC is passed to flags in tdb_open then transactions are
+    still available, but no transaction recovery area is used and no
+    fsync/msync calls are made.
+
+*/
+
+
+/*
+  hold the context of any current transaction
+*/
+struct tdb_transaction {
+	/* we keep a mirrored copy of the tdb hash heads here so
+	   tdb_next_hash_chain() can operate efficiently */
+	u32 *hash_heads;
+
+	/* the original io methods - used to do IOs to the real db */
+	const struct tdb_methods *io_methods;
+
+	/* the list of transaction elements. We use a doubly linked
+	   list with a last pointer to allow us to keep the list
+	   ordered, with first element at the front of the list. It
+	   needs to be doubly linked as the read/write traversals need
+	   to be backwards, while the commit needs to be forwards */
+	struct tdb_transaction_el {
+		struct tdb_transaction_el *next, *prev;
+		tdb_off_t offset;
+		tdb_len_t length;
+		unsigned char *data;
+	} *elements, *elements_last;
+
+	/* non-zero when an internal transaction error has
+	   occurred. All write operations will then fail until the
+	   transaction is ended */
+	int transaction_error;
+
+	/* when inside a transaction we need to keep track of any
+	   nested tdb_transaction_start() calls, as these are allowed,
+	   but don't create a new transaction */
+	int nesting;
+
+	/* old file size before transaction */
+	tdb_len_t old_map_size;
+};
+
+
+/*
+  read while in a transaction. We need to check first if the data is in our list
+  of transaction elements, then if not do a real read
+*/
+static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf, 
+			    tdb_len_t len, int cv)
+{
+	struct tdb_transaction_el *el;
+
+	/* we need to walk the list backwards to get the most recent data */
+	for (el=tdb->transaction->elements_last;el;el=el->prev) {
+		tdb_len_t partial;
+
+		if (off+len <= el->offset) {
+			continue;
+		}
+		if (off >= el->offset + el->length) {
+			continue;
+		}
+
+		/* an overlapping read - needs to be split into up to
+		   2 reads and a memcpy */
+		if (off < el->offset) {
+			partial = el->offset - off;
+			if (transaction_read(tdb, off, buf, partial, cv) != 0) {
+				goto fail;
+			}
+			len -= partial;
+			off += partial;
+			buf = (void *)(partial + (char *)buf);
+		}
+		if (off + len <= el->offset + el->length) {
+			partial = len;
+		} else {
+			partial = el->offset + el->length - off;
+		}
+		memcpy(buf, el->data + (off - el->offset), partial);
+		if (cv) {
+			tdb_convert(buf, len);
+		}
+		len -= partial;
+		off += partial;
+		buf = (void *)(partial + (char *)buf);
+		
+		if (len != 0 && transaction_read(tdb, off, buf, len, cv) != 0) {
+			goto fail;
+		}
+
+		return 0;
+	}
+
+	/* its not in the transaction elements - do a real read */
+	return tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv);
+
+fail:
+	TDB_LOG((tdb, 0, "transaction_read: failed at off=%d len=%d\n", off, len));
+	tdb->ecode = TDB_ERR_IO;
+	tdb->transaction->transaction_error = 1;
+	return -1;
+}
+
+
+/*
+  write while in a transaction
+*/
+static int transaction_write(struct tdb_context *tdb, tdb_off_t off, 
+			     const void *buf, tdb_len_t len)
+{
+	struct tdb_transaction_el *el, *best_el=NULL;
+
+	if (len == 0) {
+		return 0;
+	}
+	
+	/* if the write is to a hash head, then update the transaction
+	   hash heads */
+	if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
+	    off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
+		u32 chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
+		memcpy(&tdb->transaction->hash_heads[chain], buf, len);
+	}
+
+	/* first see if we can replace an existing entry */
+	for (el=tdb->transaction->elements_last;el;el=el->prev) {
+		tdb_len_t partial;
+
+		if (best_el == NULL && off == el->offset+el->length) {
+			best_el = el;
+		}
+
+		if (off+len <= el->offset) {
+			continue;
+		}
+		if (off >= el->offset + el->length) {
+			continue;
+		}
+
+		/* an overlapping write - needs to be split into up to
+		   2 writes and a memcpy */
+		if (off < el->offset) {
+			partial = el->offset - off;
+			if (transaction_write(tdb, off, buf, partial) != 0) {
+				goto fail;
+			}
+			len -= partial;
+			off += partial;
+			buf = (const void *)(partial + (const char *)buf);
+		}
+		if (off + len <= el->offset + el->length) {
+			partial = len;
+		} else {
+			partial = el->offset + el->length - off;
+		}
+		memcpy(el->data + (off - el->offset), buf, partial);
+		len -= partial;
+		off += partial;
+		buf = (const void *)(partial + (const char *)buf);
+		
+		if (len != 0 && transaction_write(tdb, off, buf, len) != 0) {
+			goto fail;
+		}
+
+		return 0;
+	}
+
+	/* see if we can append the new entry to an existing entry */
+	if (best_el && best_el->offset + best_el->length == off && 
+	    (off+len < tdb->transaction->old_map_size ||
+	     off > tdb->transaction->old_map_size)) {
+		unsigned char *data = best_el->data;
+		el = best_el;
+		el->data = realloc(el->data, el->length + len);
+		if (el->data == NULL) {
+			tdb->ecode = TDB_ERR_OOM;
+			tdb->transaction->transaction_error = 1;
+			el->data = data;
+			return -1;
+		}
+		if (buf) {
+			memcpy(el->data + el->length, buf, len);
+		} else {
+			memset(el->data + el->length, TDB_PAD_BYTE, len);
+		}
+		el->length += len;
+		return 0;
+	}
+
+	/* add a new entry at the end of the list */
+	el = malloc(sizeof(*el));
+	if (el == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		tdb->transaction->transaction_error = 1;		
+		return -1;
+	}
+	el->next = NULL;
+	el->prev = tdb->transaction->elements_last;
+	el->offset = off;
+	el->length = len;
+	el->data = malloc(len);
+	if (el->data == NULL) {
+		free(el);
+		tdb->ecode = TDB_ERR_OOM;
+		tdb->transaction->transaction_error = 1;		
+		return -1;
+	}
+	if (buf) {
+		memcpy(el->data, buf, len);
+	} else {
+		memset(el->data, TDB_PAD_BYTE, len);
+	}
+	if (el->prev) {
+		el->prev->next = el;
+	} else {
+		tdb->transaction->elements = el;
+	}
+	tdb->transaction->elements_last = el;
+	return 0;
+
+fail:
+	TDB_LOG((tdb, 0, "transaction_write: failed at off=%d len=%d\n", off, len));
+	tdb->ecode = TDB_ERR_IO;
+	tdb->transaction->transaction_error = 1;
+	return -1;
+}
+
+/*
+  accelerated hash chain head search, using the cached hash heads
+*/
+static void transaction_next_hash_chain(struct tdb_context *tdb, u32 *chain)
+{
+	u32 h = *chain;
+	for (;h < tdb->header.hash_size;h++) {
+		/* the +1 takes account of the freelist */
+		if (0 != tdb->transaction->hash_heads[h+1]) {
+			break;
+		}
+	}
+	(*chain) = h;
+}
+
+/*
+  out of bounds check during a transaction
+*/
+static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
+{
+	if (len <= tdb->map_size) {
+		return 0;
+	}
+	return TDB_ERRCODE(TDB_ERR_IO, -1);
+}
+
+/*
+  transaction version of tdb_expand().
+*/
+static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t size, 
+				   tdb_off_t addition)
+{
+	/* add a write to the transaction elements, so subsequent
+	   reads see the zero data */
+	if (transaction_write(tdb, size, NULL, addition) != 0) {
+		return -1;
+	}
+
+	return 0;
+}
+
+/*
+  brlock during a transaction - ignore them
+*/
+int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset, 
+		       int rw_type, int lck_type, int probe)
+{
+	return 0;
+}
+
+static const struct tdb_methods transaction_methods = {
+	transaction_read,
+	transaction_write,
+	transaction_next_hash_chain,
+	transaction_oob,
+	transaction_expand_file,
+	transaction_brlock
+};
+
+
+/*
+  start a tdb transaction. No token is returned, as only a single
+  transaction is allowed to be pending per tdb_context
+*/
+int tdb_transaction_start(struct tdb_context *tdb)
+{
+	/* some sanity checks */
+	if (tdb->read_only || (tdb->flags & TDB_INTERNAL) || tdb->traverse_read) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
+		tdb->ecode = TDB_ERR_EINVAL;
+		return -1;
+	}
+
+	/* cope with nested tdb_transaction_start() calls */
+	if (tdb->transaction != NULL) {
+		tdb->transaction->nesting++;
+		TDB_LOG((tdb, 0, "tdb_transaction_start: nesting %d\n", 
+			 tdb->transaction->nesting));
+		return 0;
+	}
+
+	if (tdb->num_locks != 0) {
+		/* the caller must not have any locks when starting a
+		   transaction as otherwise we'll be screwed by lack
+		   of nested locks in posix */
+		TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction with locks held\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->travlocks.next != NULL) {
+		/* you cannot use transactions inside a traverse (although you can use
+		   traverse inside a transaction) as otherwise you can end up with
+		   deadlock */
+		TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	tdb->transaction = calloc(sizeof(struct tdb_transaction), 1);
+	if (tdb->transaction == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	/* get the transaction write lock. This is a blocking lock. As
+	   discussed with Volker, there are a number of ways we could
+	   make this async, which we will probably do in the future */
+	if (tdb_brlock_len(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: failed to get transaction lock\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		SAFE_FREE(tdb->transaction);
+		return -1;
+	}
+	
+	/* get a read lock from the freelist to the end of file. This
+	   is upgraded to a write lock during the commit */
+	if (tdb_brlock_len(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: failed to get hash locks\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		goto fail;
+	}
+
+	/* setup a copy of the hash table heads so the hash scan in
+	   traverse can be fast */
+	tdb->transaction->hash_heads = calloc(tdb->header.hash_size+1, sizeof(tdb_off_t));
+	if (tdb->transaction->hash_heads == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		goto fail;
+	}
+	if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
+				   TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: failed to read hash heads\n"));
+		tdb->ecode = TDB_ERR_IO;
+		goto fail;
+	}
+
+	/* make sure we know about any file expansions already done by
+	   anyone else */
+	tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+	tdb->transaction->old_map_size = tdb->map_size;
+
+	/* finally hook the io methods, replacing them with
+	   transaction specific methods */
+	tdb->transaction->io_methods = tdb->methods;
+	tdb->methods = &transaction_methods;
+
+	/* by calling this transaction write here, we ensure that we don't grow the
+	   transaction linked list due to hash table updates */
+	if (transaction_write(tdb, FREELIST_TOP, tdb->transaction->hash_heads, 
+			      TDB_HASHTABLE_SIZE(tdb)) != 0) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: failed to prime hash table\n"));
+		tdb->ecode = TDB_ERR_IO;
+		goto fail;
+	}
+
+	return 0;
+	
+fail:
+	tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+	tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+	SAFE_FREE(tdb->transaction->hash_heads);
+	SAFE_FREE(tdb->transaction);
+	return -1;
+}
+
+
+/*
+  cancel the current transaction
+*/
+int tdb_transaction_cancel(struct tdb_context *tdb)
+{	
+	if (tdb->transaction == NULL) {
+		TDB_LOG((tdb, 0, "tdb_transaction_cancel: no transaction\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->nesting != 0) {
+		tdb->transaction->transaction_error = 1;
+		tdb->transaction->nesting--;
+		return 0;
+	}		
+
+	tdb->map_size = tdb->transaction->old_map_size;
+
+	/* free all the transaction elements */
+	while (tdb->transaction->elements) {
+		struct tdb_transaction_el *el = tdb->transaction->elements;
+		tdb->transaction->elements = el->next;
+		free(el->data);
+		free(el);
+	}
+
+	/* remove any locks created during the transaction */
+	if (tdb->num_locks != 0) {
+		int h;
+		for (h=0;h<tdb->header.hash_size+1;h++) {
+			if (tdb->locked[h].count != 0) {
+				tdb_brlock_len(tdb,FREELIST_TOP+4*h,F_UNLCK,F_SETLKW, 0, 1);
+				tdb->locked[h].count = 0;
+			}
+		}
+		tdb->num_locks = 0;
+	}
+
+	/* restore the normal io methods */
+	tdb->methods = tdb->transaction->io_methods;
+
+	tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+	tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+	SAFE_FREE(tdb->transaction->hash_heads);
+	SAFE_FREE(tdb->transaction);
+	
+	return 0;
+}
+
+/*
+  sync to disk
+*/
+static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
+{	
+	if (fsync(tdb->fd) != 0) {
+		tdb->ecode = TDB_ERR_IO;
+		TDB_LOG((tdb, 0, "tdb_transaction: fsync failed\n"));
+		return -1;
+	}
+#ifdef MS_SYNC
+	if (tdb->map_ptr) {
+		tdb_off_t moffset = offset & ~(tdb->page_size-1);
+		if (msync(moffset + (char *)tdb->map_ptr, 
+			  length + (offset - moffset), MS_SYNC) != 0) {
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, 0, "tdb_transaction: msync failed - %s\n",
+				 strerror(errno)));
+			return -1;
+		}
+	}
+#endif
+	return 0;
+}
+
+
+/*
+  work out how much space the linearised recovery data will consume
+*/
+static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
+{
+	struct tdb_transaction_el *el;
+	tdb_len_t recovery_size = 0;
+
+	recovery_size = sizeof(u32);
+	for (el=tdb->transaction->elements;el;el=el->next) {
+		if (el->offset >= tdb->transaction->old_map_size) {
+			continue;
+		}
+		recovery_size += 2*sizeof(tdb_off_t) + el->length;
+	}
+
+	return recovery_size;
+}
+
+/*
+  allocate the recovery area, or use an existing recovery area if it is
+  large enough
+*/
+static int tdb_recovery_allocate(struct tdb_context *tdb, 
+				 tdb_len_t *recovery_size,
+				 tdb_off_t *recovery_offset,
+				 tdb_len_t *recovery_max_size)
+{
+	struct list_struct rec;
+	const struct tdb_methods *methods = tdb->transaction->io_methods;
+	tdb_off_t recovery_head;
+
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+		TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to read recovery head\n"));
+		return -1;
+	}
+
+	rec.rec_len = 0;
+
+	if (recovery_head != 0 && 
+	    methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
+		TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to read recovery record\n"));
+		return -1;
+	}
+
+	*recovery_size = tdb_recovery_size(tdb);
+
+	if (recovery_head != 0 && *recovery_size <= rec.rec_len) {
+		/* it fits in the existing area */
+		*recovery_max_size = rec.rec_len;
+		*recovery_offset = recovery_head;
+		return 0;
+	}
+
+	/* we need to free up the old recovery area, then allocate a
+	   new one at the end of the file. Note that we cannot use
+	   tdb_allocate() to allocate the new one as that might return
+	   us an area that is being currently used (as of the start of
+	   the transaction) */
+	if (recovery_head != 0) {
+		if (tdb_free(tdb, recovery_head, &rec) == -1) {
+			TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to free previous recovery area\n"));
+			return -1;
+		}
+	}
+
+	/* the tdb_free() call might have increased the recovery size */
+	*recovery_size = tdb_recovery_size(tdb);
+
+	/* round up to a multiple of page size */
+	*recovery_max_size = TDB_ALIGN(sizeof(rec) + *recovery_size, tdb->page_size) - sizeof(rec);
+	*recovery_offset = tdb->map_size;
+	recovery_head = *recovery_offset;
+
+	if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size, 
+				     (tdb->map_size - tdb->transaction->old_map_size) +
+				     sizeof(rec) + *recovery_max_size) == -1) {
+		TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to create recovery area\n"));
+		return -1;
+	}
+
+	/* remap the file (if using mmap) */
+	methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+
+	/* we have to reset the old map size so that we don't try to expand the file
+	   again in the transaction commit, which would destroy the recovery area */
+	tdb->transaction->old_map_size = tdb->map_size;
+
+	/* write the recovery header offset and sync - we can sync without a race here
+	   as the magic ptr in the recovery record has not been set */
+	CONVERT(recovery_head);
+	if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD, 
+			       &recovery_head, sizeof(tdb_off_t)) == -1) {
+		TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to write recovery head\n"));
+		return -1;
+	}
+
+	return 0;
+}
+
+
+/*
+  setup the recovery data that will be used on a crash during commit
+*/
+static int transaction_setup_recovery(struct tdb_context *tdb, 
+				      tdb_off_t *magic_offset)
+{
+	struct tdb_transaction_el *el;
+	tdb_len_t recovery_size;
+	unsigned char *data, *p;
+	const struct tdb_methods *methods = tdb->transaction->io_methods;
+	struct list_struct *rec;
+	tdb_off_t recovery_offset, recovery_max_size;
+	tdb_off_t old_map_size = tdb->transaction->old_map_size;
+	u32 magic, tailer;
+
+	/*
+	  check that the recovery area has enough space
+	*/
+	if (tdb_recovery_allocate(tdb, &recovery_size, 
+				  &recovery_offset, &recovery_max_size) == -1) {
+		return -1;
+	}
+
+	data = malloc(recovery_size + sizeof(*rec));
+	if (data == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	rec = (struct list_struct *)data;
+	memset(rec, 0, sizeof(*rec));
+
+	rec->magic    = 0;
+	rec->data_len = recovery_size;
+	rec->rec_len  = recovery_max_size;
+	rec->key_len  = old_map_size;
+	CONVERT(rec);
+
+	/* build the recovery data into a single blob to allow us to do a single
+	   large write, which should be more efficient */
+	p = data + sizeof(*rec);
+	for (el=tdb->transaction->elements;el;el=el->next) {
+		if (el->offset >= old_map_size) {
+			continue;
+		}
+		if (el->offset + el->length > tdb->transaction->old_map_size) {
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: transaction data over new region boundary\n"));
+			free(data);
+			tdb->ecode = TDB_ERR_CORRUPT;
+			return -1;
+		}
+		memcpy(p, &el->offset, 4);
+		memcpy(p+4, &el->length, 4);
+		if (DOCONV()) {
+			tdb_convert(p, 8);
+		}
+		/* the recovery area contains the old data, not the
+		   new data, so we have to call the original tdb_read
+		   method to get it */
+		if (methods->tdb_read(tdb, el->offset, p + 8, el->length, 0) != 0) {
+			free(data);
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+		}
+		p += 8 + el->length;
+	}
+
+	/* and the tailer */
+	tailer = sizeof(*rec) + recovery_max_size;
+	memcpy(p, &tailer, 4);
+	CONVERT(p);
+
+	/* write the recovery data to the recovery area */
+	if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to write recovery data\n"));
+		free(data);
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* as we don't have ordered writes, we have to sync the recovery
+	   data before we update the magic to indicate that the recovery
+	   data is present */
+	if (transaction_sync(tdb, recovery_offset, sizeof(*rec) + recovery_size) == -1) {
+		free(data);
+		return -1;
+	}
+
+	free(data);
+
+	magic = TDB_RECOVERY_MAGIC;
+	CONVERT(magic);
+
+	*magic_offset = recovery_offset + offsetof(struct list_struct, magic);
+
+	if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to write recovery magic\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* ensure the recovery magic marker is on disk */
+	if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
+		return -1;
+	}
+
+	return 0;
+}
+
+/*
+  commit the current transaction
+*/
+int tdb_transaction_commit(struct tdb_context *tdb)
+{	
+	const struct tdb_methods *methods;
+	tdb_off_t magic_offset = 0;
+	u32 zero = 0;
+
+	if (tdb->transaction == NULL) {
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: no transaction\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->transaction_error) {
+		tdb->ecode = TDB_ERR_IO;
+		tdb_transaction_cancel(tdb);
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: transaction error pending\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->nesting != 0) {
+		tdb->transaction->nesting--;
+		return 0;
+	}		
+
+	/* check for a null transaction */
+	if (tdb->transaction->elements == NULL) {
+		tdb_transaction_cancel(tdb);
+		return 0;
+	}
+
+	methods = tdb->transaction->io_methods;
+	
+	/* if there are any locks pending then the caller has not
+	   nested their locks properly, so fail the transaction */
+	if (tdb->num_locks) {
+		tdb->ecode = TDB_ERR_LOCK;
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: locks pending on commit\n"));
+		tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	/* upgrade the main transaction lock region to a write lock */
+	if (tdb_brlock_upgrade(tdb, FREELIST_TOP, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_start: failed to upgrade hash locks\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	/* get the global lock - this prevents new users attaching to the database
+	   during the commit */
+	if (tdb_brlock_len(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to get global lock\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	if (!(tdb->flags & TDB_NOSYNC)) {
+		/* write the recovery data to the end of the file */
+		if (transaction_setup_recovery(tdb, &magic_offset) == -1) {
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to setup recovery data\n"));
+			tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+			tdb_transaction_cancel(tdb);
+			return -1;
+		}
+	}
+
+	/* expand the file to the new size if needed */
+	if (tdb->map_size != tdb->transaction->old_map_size) {
+		if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size, 
+					     tdb->map_size - 
+					     tdb->transaction->old_map_size) == -1) {
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: expansion failed\n"));
+			tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+			tdb_transaction_cancel(tdb);
+			return -1;
+		}
+		tdb->map_size = tdb->transaction->old_map_size;
+		methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+	}
+
+	/* perform all the writes */
+	while (tdb->transaction->elements) {
+		struct tdb_transaction_el *el = tdb->transaction->elements;
+
+		if (methods->tdb_write(tdb, el->offset, el->data, el->length) == -1) {
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: write failed during commit\n"));
+			
+			/* we've overwritten part of the data and
+			   possibly expanded the file, so we need to
+			   run the crash recovery code */
+			tdb->methods = methods;
+			tdb_transaction_recover(tdb); 
+
+			tdb_transaction_cancel(tdb);
+			tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: write failed\n"));
+			return -1;
+		}
+		tdb->transaction->elements = el->next;
+		free(el->data); 
+		free(el);
+	} 
+
+	if (!(tdb->flags & TDB_NOSYNC)) {
+		/* ensure the new data is on disk */
+		if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+			return -1;
+		}
+
+		/* remove the recovery marker */
+		if (methods->tdb_write(tdb, magic_offset, &zero, 4) == -1) {
+			TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to remove recovery magic\n"));
+			return -1;
+		}
+
+		/* ensure the recovery marker has been removed on disk */
+		if (transaction_sync(tdb, magic_offset, 4) == -1) {
+			return -1;
+		}
+	}
+
+	tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+
+	/*
+	  TODO: maybe write to some dummy hdr field, or write to magic
+	  offset without mmap, before the last sync, instead of the
+	  utime() call
+	*/
+
+	/* on some systems (like Linux 2.6.x) changes via mmap/msync
+	   don't change the mtime of the file, this means the file may
+	   not be backed up (as tdb rounding to block sizes means that
+	   file size changes are quite rare too). The following forces
+	   mtime changes when a transaction completes */
+#ifdef HAVE_UTIME
+	utime(tdb->name, NULL);
+#endif
+
+	/* use a transaction cancel to free memory and remove the
+	   transaction locks */
+	tdb_transaction_cancel(tdb);
+	return 0;
+}
+
+
+/*
+  recover from an aborted transaction. Must be called with exclusive
+  database write access already established (including the global
+  lock to prevent new processes attaching)
+*/
+int tdb_transaction_recover(struct tdb_context *tdb)
+{
+	tdb_off_t recovery_head, recovery_eof;
+	unsigned char *data, *p;
+	u32 zero = 0;
+	struct list_struct rec;
+
+	/* find the recovery area */
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery head\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	if (recovery_head == 0) {
+		/* we have never allocated a recovery record */
+		return 0;
+	}
+
+	/* read the recovery record */
+	if (tdb->methods->tdb_read(tdb, recovery_head, &rec, 
+				   sizeof(rec), DOCONV()) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery record\n"));		
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	if (rec.magic != TDB_RECOVERY_MAGIC) {
+		/* there is no valid recovery data */
+		return 0;
+	}
+
+	if (tdb->read_only) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: attempt to recover read only database\n"));
+		tdb->ecode = TDB_ERR_CORRUPT;
+		return -1;
+	}
+
+	recovery_eof = rec.key_len;
+
+	data = malloc(rec.data_len);
+	if (data == NULL) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to allocate recovery data\n"));		
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	/* read the full recovery data */
+	if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
+				   rec.data_len, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery data\n"));		
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* recover the file data */
+	p = data;
+	while (p+8 < data + rec.data_len) {
+		u32 ofs, len;
+		if (DOCONV()) {
+			tdb_convert(p, 8);
+		}
+		memcpy(&ofs, p, 4);
+		memcpy(&len, p+4, 4);
+
+		if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
+			free(data);
+			TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+		}
+		p += 8 + len;
+	}
+
+	free(data);
+
+	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to sync recovery\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* if the recovery area is after the recovered eof then remove it */
+	if (recovery_eof <= recovery_head) {
+		if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
+			TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to remove recovery head\n"));
+			tdb->ecode = TDB_ERR_IO;
+			return -1;			
+		}
+	}
+
+	/* remove the recovery magic */
+	if (tdb_ofs_write(tdb, recovery_head + offsetof(struct list_struct, magic), 
+			  &zero) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to remove recovery magic\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;			
+	}
+	
+	/* reduce the file size to the old size */
+	tdb_munmap(tdb);
+	if (ftruncate(tdb->fd, recovery_eof) != 0) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to reduce to recovery size\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;			
+	}
+	tdb->map_size = recovery_eof;
+	tdb_mmap(tdb);
+
+	if (transaction_sync(tdb, 0, recovery_eof) == -1) {
+		TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to sync2 recovery\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	TDB_LOG((tdb, 0, "tdb_transaction_recover: recovered %d byte database\n", 
+		 recovery_eof));
+
+	/* all done */
+	return 0;
+}
diff --git a/source3/tdb/traverse.c b/source3/tdb/traverse.c
new file mode 100644
index 0000000000..d070e0a766
--- /dev/null
+++ b/source3/tdb/traverse.c
@@ -0,0 +1,334 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include "tdb_private.h"
+
+/* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
+static int tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
+			 struct list_struct *rec)
+{
+	int want_next = (tlock->off != 0);
+
+	/* Lock each chain from the start one. */
+	for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
+		if (!tlock->off && tlock->hash != 0) {
+			/* this is an optimisation for the common case where
+			   the hash chain is empty, which is particularly
+			   common for the use of tdb with ldb, where large
+			   hashes are used. In that case we spend most of our
+			   time in tdb_brlock(), locking empty hash chains.
+			   
+			   To avoid this, we do an unlocked pre-check to see
+			   if the hash chain is empty before starting to look
+			   inside it. If it is empty then we can avoid that
+			   hash chain. If it isn't empty then we can't believe
+			   the value we get back, as we read it without a
+			   lock, so instead we get the lock and re-fetch the
+			   value below.
+			   
+			   Notice that not doing this optimisation on the
+			   first hash chain is critical. We must guarantee
+			   that we have done at least one fcntl lock at the
+			   start of a search to guarantee that memory is
+			   coherent on SMP systems. If records are added by
+			   others during the search then thats OK, and we
+			   could possibly miss those with this trick, but we
+			   could miss them anyway without this trick, so the
+			   semantics don't change.
+			   
+			   With a non-indexed ldb search this trick gains us a
+			   factor of around 80 in speed on a linux 2.6.x
+			   system (testing using ldbtest).
+			*/
+			tdb->methods->next_hash_chain(tdb, &tlock->hash);
+			if (tlock->hash == tdb->header.hash_size) {
+				continue;
+			}
+		}
+
+		if (tdb_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
+			return -1;
+
+		/* No previous record?  Start at top of chain. */
+		if (!tlock->off) {
+			if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
+				     &tlock->off) == -1)
+				goto fail;
+		} else {
+			/* Otherwise unlock the previous record. */
+			if (tdb_unlock_record(tdb, tlock->off) != 0)
+				goto fail;
+		}
+
+		if (want_next) {
+			/* We have offset of old record: grab next */
+			if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+				goto fail;
+			tlock->off = rec->next;
+		}
+
+		/* Iterate through chain */
+		while( tlock->off) {
+			tdb_off_t current;
+			if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+				goto fail;
+
+			/* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
+			if (tlock->off == rec->next) {
+				TDB_LOG((tdb, 0, "tdb_next_lock: loop detected.\n"));
+				goto fail;
+			}
+
+			if (!TDB_DEAD(rec)) {
+				/* Woohoo: we found one! */
+				if (tdb_lock_record(tdb, tlock->off) != 0)
+					goto fail;
+				return tlock->off;
+			}
+
+			/* Try to clean dead ones from old traverses */
+			current = tlock->off;
+			tlock->off = rec->next;
+			if (!(tdb->read_only || tdb->traverse_read) && 
+			    tdb_do_delete(tdb, current, rec) != 0)
+				goto fail;
+		}
+		tdb_unlock(tdb, tlock->hash, tlock->lock_rw);
+		want_next = 0;
+	}
+	/* We finished iteration without finding anything */
+	return TDB_ERRCODE(TDB_SUCCESS, 0);
+
+ fail:
+	tlock->off = 0;
+	if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
+		TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
+	return -1;
+}
+
+/* traverse the entire database - calling fn(tdb, key, data) on each element.
+   return -1 on error or the record count traversed
+   if fn is NULL then it is not called
+   a non-zero return value from fn() indicates that the traversal should stop
+  */
+static int tdb_traverse_internal(struct tdb_context *tdb, 
+				 tdb_traverse_func fn, void *private_data,
+				 struct tdb_traverse_lock *tl)
+{
+	TDB_DATA key, dbuf;
+	struct list_struct rec;
+	int ret, count = 0;
+
+	/* This was in the initializaton, above, but the IRIX compiler
+	 * did not like it.  crh
+	 */
+	tl->next = tdb->travlocks.next;
+
+	/* fcntl locks don't stack: beware traverse inside traverse */
+	tdb->travlocks.next = tl;
+
+	/* tdb_next_lock places locks on the record returned, and its chain */
+	while ((ret = tdb_next_lock(tdb, tl, &rec)) > 0) {
+		count++;
+		/* now read the full record */
+		key.dptr = tdb_alloc_read(tdb, tl->off + sizeof(rec), 
+					  rec.key_len + rec.data_len);
+		if (!key.dptr) {
+			ret = -1;
+			if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
+				goto out;
+			if (tdb_unlock_record(tdb, tl->off) != 0)
+				TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
+			goto out;
+		}
+		key.dsize = rec.key_len;
+		dbuf.dptr = key.dptr + rec.key_len;
+		dbuf.dsize = rec.data_len;
+
+		/* Drop chain lock, call out */
+		if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
+			ret = -1;
+			SAFE_FREE(key.dptr);
+			goto out;
+		}
+		if (fn && fn(tdb, key, dbuf, private_data)) {
+			/* They want us to terminate traversal */
+			ret = count;
+			if (tdb_unlock_record(tdb, tl->off) != 0) {
+				TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));;
+				ret = -1;
+			}
+			SAFE_FREE(key.dptr);
+			goto out;
+		}
+		SAFE_FREE(key.dptr);
+	}
+out:
+	tdb->travlocks.next = tl->next;
+	if (ret < 0)
+		return -1;
+	else
+		return count;
+}
+
+
+/*
+  a write style traverse - temporarily marks the db read only
+*/
+int tdb_traverse_read(struct tdb_context *tdb, 
+		      tdb_traverse_func fn, void *private_data)
+{
+	struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
+	int ret;
+	
+	/* we need to get a read lock on the transaction lock here to
+	   cope with the lock ordering semantics of solaris10 */
+	if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_RDLCK, F_SETLKW, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_traverse_read: failed to get transaction lock\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	tdb->traverse_read++;
+	ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+	tdb->traverse_read--;
+
+	tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0);
+
+	return ret;
+}
+
+/*
+  a write style traverse - needs to get the transaction lock to
+  prevent deadlocks
+*/
+int tdb_traverse(struct tdb_context *tdb, 
+		 tdb_traverse_func fn, void *private_data)
+{
+	struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		return tdb_traverse_read(tdb, fn, private_data);
+	}
+	
+	if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
+		TDB_LOG((tdb, 0, "tdb_traverse: failed to get transaction lock\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+
+	tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0);
+
+	return ret;
+}
+
+
+/* find the first entry in the database and return its key */
+TDB_DATA tdb_firstkey(struct tdb_context *tdb)
+{
+	TDB_DATA key;
+	struct list_struct rec;
+
+	/* release any old lock */
+	if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
+		return tdb_null;
+	tdb->travlocks.off = tdb->travlocks.hash = 0;
+
+	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
+		return tdb_null;
+	/* now read the key */
+	key.dsize = rec.key_len;
+	key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
+	if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
+		TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
+	return key;
+}
+
+/* find the next entry in the database, returning its key */
+TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
+{
+	u32 oldhash;
+	TDB_DATA key = tdb_null;
+	struct list_struct rec;
+	char *k = NULL;
+
+	/* Is locked key the old key?  If so, traverse will be reliable. */
+	if (tdb->travlocks.off) {
+		if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
+			return tdb_null;
+		if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
+		    || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
+					    rec.key_len))
+		    || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
+			/* No, it wasn't: unlock it and start from scratch */
+			if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
+				SAFE_FREE(k);
+				return tdb_null;
+			}
+			if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0) {
+				SAFE_FREE(k);
+				return tdb_null;
+			}
+			tdb->travlocks.off = 0;
+		}
+
+		SAFE_FREE(k);
+	}
+
+	if (!tdb->travlocks.off) {
+		/* No previous element: do normal find, and lock record */
+		tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), F_WRLCK, &rec);
+		if (!tdb->travlocks.off)
+			return tdb_null;
+		tdb->travlocks.hash = BUCKET(rec.full_hash);
+		if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
+			TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
+			return tdb_null;
+		}
+	}
+	oldhash = tdb->travlocks.hash;
+
+	/* Grab next record: locks chain and returned record,
+	   unlocks old record */
+	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
+		key.dsize = rec.key_len;
+		key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
+					  key.dsize);
+		/* Unlock the chain of this new record */
+		if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
+			TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+	}
+	/* Unlock the chain of old record */
+	if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
+		TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+	return key;
+}
+